Interactive comment on “ Air pollution and associated human mortality : the role of air pollutant emissions , climate change and methane concentration increases during the industrial period

Introduction Conclusions References

climate change and atmospheric chemistry has influenced atmospheric composition and human mortality associated with industrial air pollution.In addition to driving 13 % of the total historical changes in surface O 3 and 15 % of the associated mortalities, CH 4 is the dominant factor driving changes in atmospheric OH and H 2 O 2 since preindustrial time.Our study highlights the benefits to air quality and human health of CH 4 mitigation as a component of future air pollution control policy.

Introduction
Human activities since preindustrial time have resulted in large increases in air pollution (IPCC, 2001).Measurements at various sites in the Northern Hemisphere indicate an increase from the 1860s to 2000s in surface ozone (O 3 ) of approximately a factor of 4 (from about 10 to 50 ppbv) (Gros, 2006;Marenco et al., 1994).Sulfate and carbonaceous aerosol concentrations in Greenland ice cores suggest a factor of 3-4 increase from the mid-1860s to the present (D öscher et al., 1995;Fischer et al., 1998;Lavanchy et al., 1999).Sulfate and carbonaceous aerosols are key components of fine particulate matter (≤ 2.5 µm aerodynamic diameter, PM 2.5 ) which, along with O 3 , are pollutants that adversely impact human health (Jerrett et al., 2009;Krewski et al., 2009;Pope and Dockery, 2006;Levy et al., 2005;Bell et al., 2004;Pope et al., 2002).
Here, we apply simulations of a global coupled chemistry-climate model to investigate changes in O 3 and PM 2.5 from the preindustrial era to the present and their associated effects on premature mortality.
O 3 is a secondary air pollutant that is formed in the troposphere by catalytic photochemical reactions of nitrogen oxides (NO x = NO + NO 2 ) with carbon monoxide (CO), methane (CH 4 ) and other volatile organic compounds (VOCs).PM 2.5 , including sulfate, nitrate, Organic Carbon (OC), Black Carbon (BC), secondary organic carbon, fine dust and sea salt, is either directly emitted from various sources or produced via chemical reactions between directly-emitted gas-phase precursors (including SO 2 , NO x , NH 3 biogenic VOCs etc. and PM 2.5 concentrations over this period  are difficult to quantify because of sparse and uncertain preindustrial measurements, spatial heterogeneity of these species, uncertainties in estimating preindustrial emissions, and the non-linear dependence of O 3 and PM 2.5 on their precursor emissions (Horowitz, 2006).Changes in surface O 3 and PM 2.5 concentrations are largely controlled by changes in emissions of their precursors.Consequently, many recent studies have applied Chemical Transport Models (CTMs) to estimate changes in tropospheric O 3 and aerosol concentrations from the preindustrial era to the present (Horowitz, 2006;Tsigaridis et al., 2006;Lamarque et al., 2005;Grenfell et al., 2001;Mickley et al., 2001;Wang and Jacob, 1998).Annenberg et al. (2010) used preindustrial and present simulations from one of these CTM modeling studies (Horowitz 2006) to estimate the effect of anthropogenic O 3 and PM 2.5 on present premature human mortality.However, these studies, which usually apply different emissions of short-lived species but use the same meteorological driver for preindustrial and present day simulations, do not take into account the interaction between climate and air pollution (Jacob and Winner, 2009;Isaksen et al., 2009;Fiore et al., 2012).Some short-lived species are radiatively active; therefore, they perturb climate and meteorology from regional to global scales (Naik et al., 2012;Levy et al., 2008;Shindell et al., 2008).As a result, quantifying the impact of their emission changes on air quality using CTM simulations driven by the same meteorology neglects the feedbacks between short-lived species and climate.Conversely, studies have shown that climate change can affect surface O 3 and PM 2.5 concentrations and thus indirectly affect human mortality (Fang et al., 2012;Tagaris et al., 2009;Bell et al., 2007).Additionally, methane (CH 4 ) concentration changes (from 800 ppbv in 1860 to 1750 ppbv in 2000) not only give a direct radiative forcing of +0.42 W m −2 (calculated as in Ramaswamy et al., 2001), but also contribute to increasing O 3 concentrations.To understand changes in surface O 3 and PM 2.5 over the industrial period (defined here as , we need to consider the effects of changing emissions of short-lived species, climate and CH 4 concentrations on surface air quality and allow feedbacks between chemistry and climate to take place.Introduction

Conclusions References
Tables Figures

Back Close
Full In this paper, we utilize the Geophysical Fluid Dynamics Laboratory (GFDL) Atmospheric Model, version 3 (AM3), a newly developed global 3-D model that fully couples atmospheric chemistry and climate.Our goal is to understand changes in O 3 and PM 2.5 from the preindustrial era to the present ("industrial" or "historic" period) and their associated effects on premature mortality.We further attribute the changing PM 2.5 and O 3 concentrations over this period to three factors: (1) changes in direct emissions of their constituents and precursors; (2) climate change induced changes in surface concentrations, and (3) the influence of increasing methane (CH 4 ) concentrations on tropospheric chemistry.For each factor we estimate the associated impact on human health.The GFDL AM3 model and our simulations are described in Sect. 2. We evaluate simulated surface O 3 and PM 2.5 concentrations in Sect.3. Changes in surface air quality are attributed to specific factors in Sect. 4. In Sect.5, we calculate the changes in premature mortality associated with the simulated changes in air quality.Findings and conclusions are presented in Sect.6.

Model description
The AM3 model (Donner et al., 2011)  grid with the grid size varying from 163 km (at the corners of each face) to 231 km (near the center of each face).Vertically, the model extends from the surface up to 0.01 hPa (86 km) with 48 vertical hybrid sigma pressure levels.

Simulations
To investigate the change in concentrations of O 3 and PM 2.5 during the "industrial period", we use AM3 timeslice simulations -"1860" and "2000".Results from these simulations were contributed to the Atmospheric Chemistry and Climate Model Intercomparison Project (ACCMIP) (Lamarque et al., 2012).To evaluate the relative importance of the drivers of air pollutant concentration changes: (1) anthropogenic and biomass burning emissions of short-lived air pollutants, (2) climate and (3) CH 4 concentration, we also analyze three additional sensitivity simulations.The five simulations used are summarized in Table 1 and described briefly below.We use the "1860" and "2000" simulations to quantify the change in air quality during the "industrial period".These simulations use prescribed mean climatological Sea Surface Temperature (SST) and Sea Ice Cover (SIC) for the decade 1860-1869 and 1995-2004, respectively, taken from one member of the 5-member ensemble historical simulation of the GFDL CM3 model conducted in support of the Intergovernmental Panel on Climate Change-Fifth Assessment Report (IPCC-AR5).Well-mixed greenhouse gases (WMGG), including CO 2 , N 2 O, CH 4 and CFCs, are specified for the years 1860 and 2000 according to the database (http://www.iiasa.ac.at/web-apps/tnt/RcpDb/dsd?Action=htmlpage&page=welcome) developed in support of the IPCC-AR5 (Meinshausen et al., 2011).Global mean CH 4 concentration is specified at the surface at 1860 and 2000 levels, respectively, as lower boundary conditions for tropospheric chemistry calculations Anthropogenic emissions (including from energy production, industry, land transport, maritime transport, aviation, residential and commercial sectors, solvents, agriculture, agriculture waste burning on fields, and waste) and biomass burning emissions (including from open vegetation fires in forests, savanna and grasslands) of short-lived air pollutants (i.e., NO x , CO, SO 2 , NMVOCs, black carbon and Introduction

Conclusions References
Tables Figures

Back Close
Full organic carbon, etc.) for 1860 and 2000 are from the emission inventory of Lamarque et al. (2010).Anthropogenic emissions for 2000 were constructed by aggregating existing regional and global emission inventories for 40 world regions and 10 sectors.Anthropogenic emissions for 1860 were generated based on extrapolation of the EDGAR-HYDE emission inventory from 1890 to 1850 using global fossil fuel consumption estimates from Andres et al. (1999) and regional scale data for population from the HYDE dataset (Goldewijk, 2005).Biomass burning emissions in 2000 are from the GFED2 emission inventory (van der Werf et al., 2006); the 1900-2000 biomass burning emission trend is taken from RETRO (1960-2000, Schultz et al., 2008) and GICC (1900( -1950( , Mieville et al., 2010) ) inventories; no trend is assumed between 1850 and 1900 as suggested by ice-core and charcoal records (Marlon et al., 2008;McConnell et al., 2007).The 1860 and 2000 anthropogenic and biomass burning emissions used are summarized in Table S1 of the Supplement.As described by Naik et al. (2012), all natural emissions except those that depend on the simulated meteorology (lightning NO x dimethyl sulfide (DMS), sea salt and dust emissions), are the same for both simulations.Ozone depleting substances (ODS) are set to pre-1950 levels in the "1860" simulation and to 2000 levels in the "2000" simulation.All simulations were run for 11 yr with the first year used as spin up.
We define the difference in PM 2.5 and O 3 concentrations between the AM3 "2000" and "1860" simulations as "industrial" air pollution, which reflects the total changes in air pollution levels between the start of the industrial period and 2000.Our definition of "industrial" pollution includes not only the effect of changes in short-lived air pollutant emissions, but also the effect of changes in climate, CH 4 and ODS concentrations.It differs from the "anthropogenic" pollution defined in Anenberg et al. (2010), which was estimated as the difference between two CTM simulations with different emissions and, therefore, solely reflected the impact of emission changes on air pollution in a system that did not allow feedbacks between chemistry and climate.
To isolate the individual impacts of changes in emissions, climate, and CH 4 concentrations on surface concentrations of O 3 and PM 2.5 , we analyze three additional AM3 Introduction

Conclusions References
Tables Figures

Back Close
Full simulations (simulations 3, 4 and 5 in Table 1).Briefly, simulation "2000CL1860EM" is similar to the "2000" simulation but uses 1860 emissions of short-lived species; simulation "1860CL2000EM" is also similar to the "2000" simulation but its' SST, SIC, WMGG and ODS concentrations (applied in stratospheric chemistry and radiative forcing calculations) are set to 1860 levels; simulation "1860ALL2000EM" is similar to "1860CL2000EM", but its global mean CH 4 concentration is also specified at 1860 levels for tropospheric chemistry calculations.Impacts of (1) changing emissions of shortlived species, (2) changing climate and (3) changing CH 4 concentrations on air quality are estimated in our study as "2000"-"2000CL1860EM", "2000"-"1860CL2000EM", and "1860CL2000EM"-"1860ALL2000EM", respectively.In order to distinguish signals driven by changing emissions, CH 4 and climate from internal model variability, we use annually-invariant SST, SIC and air pollutant emissions to drive 10-yr model simulations and analyze averages of these 10-yr simulations.To indicate the significance of a result relative to the inter-annual variability due to internal model variability, we report a mean (average of all 10 yr) along with the standard deviation (root mean square of variance over the 10 yr) in the following sections (as mean ± std).

Adverse health impacts
We analyze the effect of industrial air pollution on premature mortality using health impact functions that relate changes in air pollutant concentrations to changes in mortality.We further evaluate the relative importance of changes in emissions of air pollutants, climate change and increased CH 4 concentrations on the incidence of premature mortalities associated with air pollution.
To obtain estimates of the excess mortalities (∆Mort) attributable to air pollution changes during the industrial period we use health impact functions for O 3 and PM 2.5 .These functions are based on log-linear relationships between relative risk and concentration derived from American Cancer Society (ACS) cohort studies for adults aged Introduction

Conclusions References
Tables Figures

Back Close
Full 30 and older (Jerrett et al., 2009;Krewski et al., 2009;Pope et al., 2002).We apply in each of the AM3 surface grid cells using the corresponding population (POP), baseline mortality (Mort base ), and consistent with the ACS study, the fraction of the population (Frac) ≥ 30 yr of age and the appropriate concentration-mortality response factor (β). Changes in O 3 and PM 2.5 concentrations (∆C) from specific factors (changes in short-lived emissions, climate and CH 4 ) from preindustrial to present day are obtained as the exposure indicators from the difference between two simulations as described in Sect.2.2, consistent with corresponding epidemiological studies.Table 2 summarizes the epidemiological studies and their β values that are applied in our study.We use population (CIESIN, 2005), population fraction of adults aged 30 and older (WHO, 2003) and baseline mortality for the year 2000 for all-cause mortality and for respiratory mortality (WHO, 2003) as in Fang et al. (2012).For PM 2.5 , although relative risk values for ischemic heart disease (a subset of cardiopulmonary disease) is reported and is used by Anenberg et al. (2010), we choose to use β for all-cause mortality as reporting of total mortality in developing countries is likely more reliable than attribution of death to specific diseases.We assume the ACS cohort studies conducted in the United States are valid globally, as relative risks characterized in US based time series studies are similar to those found in various studies in Europe (Levy et al., 2005) and Asia (HEI, 2010) and no cohort studies have yet been conducted in developing countries However, we recognize that there exist differences in PM 2.5 composition, health status, lifestyle, age structure, and medical attention available around the world, which could significantly affect our results Introduction

Conclusions References
Tables Figures

Back Close
Full AM3 simulation to reproduce observed surface O 3 and PM 2.5 concentrations around the world in order to increase confidence in our estimates of the excess mortalities attributable to air pollution.We evaluate surface O 3 in our 2000 simulation against surface O 3 data from observational networks in the United States (Clean Air Status and Trends Network-CASTNet) and in Europe (European Monitoring and Evaluation Programme-EMEP).Sites in these two networks are well-suited for evaluating global models as they represent regional O 3 concentrations rather than urban plumes.A comparison between the simulated and observed seasonal cycle of O 3 over twelve regions spanning the US and Europe is shown in Fig. 1.In general, like in Naik et al. (2012), the AM3 "2000" simulation reproduces the observed seasonal cycle of surface O 3 concentrations over most regions (r > 0.7 and bias < 20 ppbv, except in the Northwest United States).Naik et al. (2012) attribute the strong positive bias of O 3 in the Northwest U.S. to the model failure to capture the influence of maritime air masses on sites close to the ocean.However, over populated areas, such as the Eastern United States, South California and Europe, surface O 3 bias is relatively small (ranging from 4 ppbv to about 10 ppbv) and the simulated O 3 seasonal cycle follows observed cycles with correlation coefficients greater than 0.7.Correlation coefficients are greater than 0.95 over all 6 European regions.
The optical characteristics of aerosols simulated in AM3 were evaluated by Donner et al. (2011), who found that simulated AOD was within a factor of 2 of AERONET observations.However, annual mean surface PM 2.5 mass concentration simulated by AM3, which is directly associated with human mortality responses, has not yet been evaluated.Here, we evaluate annual mean PM 2.5 concentrations (or its major component, sulfate, where long-term PM  (2011).Figure 2 shows a consistent underestimate of total PM 2.5 over the United States.However, the bias for dust (Ginoux, in prep.) and sulfate over Europe and East Asia is much smaller (Fig. 2a).The greater underestimate of total PM 2.5 than dust or sulfate in the AM3 2000 simulation is likely related to the simulation of secondary organic aerosol (SOA).AM3 SOA production is scaled directly from terpene emissions and butane oxidation (Donner et al., 2011); therefore, it may underestimate SOA away from terpene and butane sources.Additionally, AM3 does not consider SOA produced by oxidation of isoprene and does not include in-cloud mechanisms for SOA production and thus likely underestimates total SOA production.Despite the systematic PM 2.5 underestimate of more than 20 %, AM3 captures the global spatial distribution of PM 2.5 well, with a correlation coefficient above 0.9.

Surface PM 2.5 and O 3 concentration changes
We first quantify the increase in surface concentrations of PM 2.   3c and 3d, reflect changes in pollutant concentrations resulting from changes in all factors between 1860 and 2000 (TOTAL).During industrialization, concentrations of air pollution increased over most regions, particularly over populated areas in the Northern Hemisphere.Global population-weighted annual mean PM 2.5 and H-O 3 increased from 1860 to 2000 by 8 ± 0.16 µg m −3 and 30 ± 0.16 ppbv, respectively (Table 3).Because of the coincidence of large concentration increases and densely populated areas, population-weighted global mean concentration changes are substantially higher than their land-only area-weighted mean values (3 ± 0.10 µg m −3 and 20 ± 0.12 ppbv).Increases in PM 2.5 and O 3 during the industrial period account for 61 % and 49 %, respectively, of their year 2000 concentrations.
Our simulated population-weighted changes in PM 2.  et al., 2000) in four geopolitical regions (Horowitz, 2006).Their preindustrial fossil fuel emissions are assumed to be zero, while their preindustrial emissions from burning of biofuels, savannah, tropical forest and agricultural waste are assumed to be 10 % of their standard 1990 emissions.Preindustrial emissions of NO x , CO, BC and OC in Introduction

Conclusions References
Tables Figures

Back Close
Full  From preindustrial time to present day, simulated tropospheric air mass-weighted OH and H 2 O 2 concentrations change by −0.11 ± 0.01 × 10 6 and +0.84 ± 0.01 × 10 10 molec cm −3 , approximately −8 % and +75 % relative to their 1860 levels.The simulated decrease in OH and increase in H 2 O 2 concentrations from the preindustrial to present period are consistent with observational records from Greenland ice cores and permafrost, respectively (Sigg and Neftel, 1991;Staffelbach et al., 1991).Changes in OH concentrations are also within the range of estimates in the literature (Naik et al., 2012;John et al., 2012).
In the following subsections, we separately explore the role of historical changes in emissions of short-lived species, climate and CH 4 on surface PM 2.5 , O 3 , and atmospheric OH and H 2 O 2 , using sensitivity simulations introduced in Sect.2.2.Introduction

Conclusions References
Tables Figures

Back Close
Full

Impact of changes in emissions of short-lived species
Here we quantify how historical changes in short-lived species emissions affect surface air quality by comparing the "2000" simulation with a simulation with 2000 climate (SST, SIC and WMGG) but 1860 emissions of short-lived species ("2000CL1860EM").Global CH 4 concentrations applied as the lower boundary condition for tropospheric chemistry calculations are specified at 2000 levels in both simulations.The same simulations have been analyzed by Naik et al. (2012) to evaluate the effect of increased emissions of short-lived species on tropospheric composition and climate forcing.In contrast, we analyze changes in surface PM 2.5 and O 3 , two species of major concern to public health.
Distributions of surface PM 2.5 and H-O 3 changes resulting from changes in shortlived air pollutant emissions are shown in Figs.3e and 3f ("2000" -"2000CL1860EM").Changing emissions raises the global annual mean population-weighted concentration of PM 2.5 by 7.5 ± 0.19 µg m −3 (Table 3), accounting for 94 % of the total change in PM 2.5 from 1860 to 2000.The PM 2.5 pattern resulting from changes in short-lived pollutant emissions (Figure 3e) is strongly correlated (R = 0.99) with the pattern of the total changes resulting from all three factors between 1860 and 2000 (Fig. 3c).
Emission changes applied in this study include both anthropogenic and biomass burning emissions, both of which were influenced by human activity during the industrial period.Anthropogenic emissions increase during the industrial period almost everywhere while changes in biomass burning emissions driven by human activities are spatially inhomogeneous.For example, from 1860 to 2000, sulfate concentrations increase everywhere, especially in the Northern Hemisphere mid-latitudes, driven by enhanced anthropogenic precursor emissions (a factor of 18 increase in sulfur dioxide emissions, Table 2 in Naik et al., 2012).However, BC and OC decrease over the U.S. while increasing over Central Africa and tropical South America.This is driven by less biomass burning in the United States and more in the tropics in 2000 than in 1860 (Fig. 6  surface area and biomass burning emissions have decreased over the past one to two centuries as a result of human-induced land-use change and fire suppression policy.In contrast, over Central Africa, Central South America and Indonesia, biomass burning emissions have increased over the past few decades due to human pressure and to the use of fire for deforestation as part of agriculture expansion (Mieville et al., 2010).Naik et al. (2012) found that increases in O 3 precursor emissions from preindustrial time to the present lead to 45 % and 40 % increases in the photochemical production and loss of O 3 , respectively, leading to a 21 % increase in tropospheric O 3 burden.As a result, the global mean population-weighted H-O 3 increases by 25 ± 0.30 pbbv (Table 3).Although the distribution of emission-driven changes in H-O 3 (Fig. 3f) correlates well with that of the total change in H-O 3 from 1860 to 2000 (Fig. 3d, R = 0.99), the emission-driven increase of H-O 3 accounts for only 83 % of the total increase (30 ± 0.16 ppbv), indicating that other factors also influence changes in H-O 3 concentrations.We show in Sects.4.3 and 4.4 that changing climate and increased CH 4 concentrations drive the rest of the H-O 3 increase.
OH and H 2 O 2 concentrations are also affected by increases in emissions of shortlived species.Tropospheric air mass-weighted OH and H 2 O 2 concentrations increase from 1860 to 2000 by +0.13 ± 0.01 × 10 6 and +0.41 ± 0.02 × 10 10 molec cm −3 due to changes in short-lived emissions alone, about +10 % and +37 % relative to their 1860 values, respectively.Increases of OH and H 2 O 2 , combined with higher SO 2 emissions, further increase sulfate production.However these short-lived species emissiondriven changes are substantially different from the total changes in OH and H 2 O 2 (−0.11 ± 0.01 × 10 6 and +0.84 ± 0.01 × 10 10 molec cm −3 ), indicating that other factors strongly influence their concentrations as shown in the following subsections.Due to increases in emissions of short-lived species, aerosols become more abundant in the atmosphere and are associated with decreased precipitation in the AM3 model (Naik et al., 2012;Donner et al., 2011).In our study, due to feedbacks resulting from increases in emissions of short-lived species, global precipitation decreases by 0.03 mm day −1 with significant large decreases (about 1mm day −1 ) over Introduction

Conclusions References
Tables Figures

Back Close
Full source regions, such as East Asia; and as a result, sulfate lifetime to wet deposition increases from 6.4 days in 1860 to 6.9 days in 2000.Increases in emissions of short-lived species also cause a net negative all-sky radiative forcing (−1.43 W m −2 ) (Naik et al., 2012), cooling the atmosphere and reducing chemical reaction rates (although this effect varies locally depending on the location and type of aerosols).Our study using AM3 captures these chemistry-climate feedback processes associated with changes in aerosol/trace gas abundance in the atmosphere and goes beyond previous CTM studies that investigate the impact of historical emission changes on atmospheric composition (Horowitz, 2006;Lamarque et al., 2005).

Impact of historical climate change
Historical climate change indirectly affects surface PM 2.5 and O 3 by modifying transport patterns, precipitation, water vapor, and temperature-dependent reaction rates.
Here we estimate the effect of historical climate change on surface PM 2.5 and O 3 by comparing the "2000" simulation with short-lived species emissions set at 2000 levels but with year 1860 SST, SIC and WMGG ("1860CL2000EM").The global CH 4 concentration applied as the lower boundary condition for tropospheric chemistry calculations is set to 2000 levels in both simulations so that we can isolate the role of climate change on air quality rather than the impact of CH 4 on tropospheric chemistry.The two simulations analyzed here have different ODS concentrations (Table 1), therefore, the impact of historical climate change also includes the effects of stratospheric O 3 depletion caused by anthropogenic ODS emissions over this period.However, our historical climate change does not include the climate change induced effect on biogenic emissions of air pollutant precursors, such as isoprene and terpenes.If a biogenic response to the changing climate were included, it would likely increase our estimate of changes in PM 2.5 and O 3 concentrations described below.Introduction

Conclusions References
Tables Figures

Back Close
Full From 1860 to 2000, our simulation indicates that climate change caused global population-weighted PM 2.5 concentrations to increase by 0.4 ± 0.17 µg m −3 (Table 3), accounting for 5 % of the total increase over this period.The spatial distribution of PM 2.5 changes, driven by climate change and driven by all factors, is only loosely correlated (R = 0.3).Changes in PM 2.5 driven by climate change show a complicated pattern (Fig. 3g): modest increases (up to 2 µg m −3 ) occur over East Asia, South and South-East Asia, West Africa and Central Europe; significant decreases (up to 1.5 µg m −3 ) occur over the United States, Central Asia, and Central and Western Australia.Significant decreases over Central and Western Australia are driven by lower concentrations of sea salt and dust (not shown), two PM 2.5 components with emissions dependent on meteorology.Increases that are statistically significant over Asia, India and Central Europe are driven by increases in BC, sulfate and OC, with OC accounting for a majority of the enhanced PM 2.5 .Increases in the concentration of these species is driven by reductions in large-scale precipitation, the major driver of wet deposition (Fig. 4).
Increases also result from higher H 2 O 2 concentrations (discussed below) that increase the aqueous phase production of sulfate.Decreases in PM 2.5 over the United States and Central Asia are driven by stronger large-scale precipitation/wet deposition close to their PM 2.5 precursor source regions (Fig. 4).While changes in PM 2.5 driven by climate change are statistically significant over most continental regions, 10-yr simulations are likely too short to generate a robust signal for precipitation distinguishable from noise on a regional scale (Naik et al., 2012).Figure 4 shows that simulated differences in large-scale precipitation is significant at the 95 % confidence level over part of East Asia, North India and Central Europe and many ocean regions, but is not significant elsewhere.If we lower the confidence level to 90 %, regions with statistically significant differences in precipitation become larger over Asia and India (see Supplement, Fig. S1).Introduction

Conclusions References
Tables Figures

Back Close
Full

Ozone
As a result of historical climate change and stratospheric O 3 depletion, the global population-weighted H-O 3 concentration increases by 0.5 ± 0.28 ppbv (Table 3), accounting for less than 2 % of the total historical H-O 3 change.Despite the small contribution, the spatial distribution of H-O 3 changes driven by climate change follows that driven by all factors with a correlation coefficient of 0.71.Historical climate change leads to increases in O 3 over polluted areas and at high-latitudes in the Northern Hemisphere, while decreasing O 3 over remote regions and oceans (Fig. 3h).This pattern, similar to that of surface O 3 responses to a warming climate in the future (Fang et al., 2012;Liao et al., 2006;Murazaki and Hess, 2006), is mainly associated with the tropospheric chemistry of O 3 under a warmer and wetter atmosphere.O 3 decreases over remote oceanic regions is largely driven by an increase in water vapor (total column water vapor increases by 3 % from preindustrial times), which leads to increases in HO x (HO x = OH + HO 2 ) concentrations (+3.6 ± 0.1× 10 6 molec cm −3 ).Reaction with HO x is the primary sink of O 3 at low NO x concentrations.However, surface O 3 increases by up to 3 ppbv over regions with large NO x emissions such as South China, North India, Northeast United States, Central Europe and Central Africa.(3) increased lightning that increases production of lightning NO x (+3 %).While lightning NO x is mostly formed in the free troposphere and contributes to stronger tropospheric O 3 production, its contribution to surface O 3 is small, most likely over regions where subsidence brings lightning-affected air masses to the surface, such as the Southwestern United States (Fang et al., 2010).Introduction

Conclusions References
Tables Figures

Back Close
Full During this historical period, anthropogenic emissions of ODS caused stratospheric O 3 depletion.Stratospheric O 3 depletion and the associated changes in J-values in the troposphere, combined with changes in transport between stratosphere and troposphere, also affect tropospheric O 3 .To estimate this effect on surface O 3 , we first evaluate changes in the cross-tropopause O 3 fluxes, and find the net flux decreases by almost 15 % relative to its level in the 1860 simulation.To estimate the stratospheric O 3 contribution to surface O 3 , we further analyze an additional tracer of stratospheric O 3 implemented in these simulations (O 3S ).This tracer, as described in Lin et al. (2012), is defined as O 3 above the World Meteorological Organization (WMO) thermal tropopause at each model time step.Once mixed into tropospheric air, O 3S is subject to the same transport and loss as tropospheric O 3 .Surface O 3S shows an annual global mean reduction of 0.7 ppbv, resulting from both stronger O 3 destruction and stratospheric O 3 depletion.Surface O 3S change has distinctive seasonal variations, with smaller decreases (−0.2 ppbv) in winter than in summer (−1.3 ppbv), consistent with projections that stratospheric-tropospheric exchange of O 3 increases in winter months in a warming climate (Liao et al., 2006;Collins et al., 2003;Zeng and Pyle, 2003).The decrease in O 3S concentration at the surface suggests that historical changes in stratospheric O 3 tend to decrease surface O 3 concentrations.However, this surface O 3S change likely overestimates changes in the stratospheric contribution to surface O 3 since any O 3 above the thermal tropopause is instantly labeled as "stratospheric" regardless of its actual origin (Lin et al., 2012).

H 2 O 2 and OH
We find climate change drives changes in tropospheric air mass-weighted OH and H 2 O 2 concentrations by −0.02 ± 0.01× 10 6 and +0.15 ± 0.03× 10 10 molec cm −3 (Table 3).The change in OH is small, reflecting competing effects between increases in tropospheric NO x (due to stronger lightning sources), SO 2 (due to weaker wet deposition) and water vapor (due to higher temperature), and decreases in O 3 burden.The Introduction

Conclusions References
Tables Figures

Back Close
Full increase in H 2 O 2 suggests enhanced conversion of HO 2 to H 2 O 2 rather than cycling back to OH.

Impact of increased CH 4 concentration
The simulated change in H-O 3 and PM 2.5 due to climate change explored in Sect.4.2 does not allow changes in CH 4 concentration to affect tropospheric chemistry (Table 1).Although this configuration intentionally isolates the role of climate change, increases in CH 4 concentrations from 1860 to 2000 also affect tropospheric chemistry and hence surface O 3 and PM 2.5 concentrations.We estimate the effect of increased CH 4 concentrations on O 3 and PM 2.5 concentrations by taking the difference between the "1860CL2000EM" and "1860ALL2000EM" simulations (Figs.3i and 3j).Both simulations are identical except that the global mean CH 4 concentrations applied in the tropospheric chemistry calculations are set to year 2000 (1750 ppbv) and 1860 (800 ppbv) levels, respectively.Tropospheric CH 4 perturbs the concentrations of oxidizing agents in the atmosphere, which in turn affect PM 2.5 and O 3 concentrations.In addition, CH 4 is a precursor of tropospheric ozone.Reaction with CH 4 is a primary sink of atmospheric OH.Higher CH 4 concentrations in 2000 than 1860 result in an OH decrease of 0.24 ± 0.01× 10 6 molec cm −3 and an increase of 0.35 ± 0.03× 10 10 molec cm −3 in H 2 O 2 (Table 3).As OH and H 2 O 2 are associated with the gas-phase and in-cloud production of sulfate, changing CH 4 thus indirectly influences PM 2.5 .Compensating changes in OH and H 2 O 2 lead to a small and insignificant global change in PM 2.5 (global population-weighted PM 2.5 decreases by 0.04 ± 0.24 µg m −3 , Table 3).The spatial pattern of PM 2.5 changes driven by the impact of increased CH 4 concentration is also not correlated with its total change during the industrial period.distribution of surface O 3 enhancement driven by increased CH 4 is significant everywhere in the world and is approximately 5-10 ppbv in the Northern Hemisphere and 2-5 ppbv in the Southern Hemisphere (Fig. 3j).A spatial correlation of 0.7 between changes in surface H-O 3 driven by increased CH 4 and that driven by all factors supports total O 3 changes being partly driven by CH 4 .Although the impact of CH 4 on O 3 has been discussed in previous literature (Fiore et al., 2008(Fiore et al., , 2002;;West et al., 2006;Dentener et al., 2005), most recent studies focus on exploring the potential benefit of future CH 4 mitigation while our study examines the total change in O 3 resulting from historic increases in CH 4 .Fiore et al. (2008), however, estimated that anthropogenic CH 4 contributes 5 ppbv to global mean surface O 3 which, despite differences in approach, is consistent with our result of a 4.3 ppbv increase in H-O 3 .
5 Premature mortalities associated with industrial air pollution

Estimate of premature mortalities associated with industrial air pollution
We estimate excess mortalities attributable to industrial air pollution separately for O 3 and PM 2.5 , using population and baseline mortality rates at present (2000) along with concentration changes in O 3 and PM 2.5 from 1860 to 2000.Globally, in 2000, industrial PM 2.5 is associated with 1.52 (95 % confidence interval, CI, of 1.03-1.98) million allcause mortalities per year (Table 4); while industrial O 3 is associated with 0.37 (95 % CI, 0.13-0.59) million respiratory mortalities per year (Table 4).Our estimates suggest that about 1.5 million all-cause premature mortalities and about 0. in Anenberg et al. (2010) (33 % and 28 %, respectively for mortalities associated with PM 2.5 and O 3 ) because our preindustrial emissions and hence simulated preindustrial O 3 and PM 2.5 are higher than theirs (see Sect. 4 and Table S1).As differences with and without use of the LCT are relatively small in our study, we hereafter only report mortalities without the LCT.
We separate the world into 10 regions to estimate the regional mortalities associated with industrial air pollution.The distribution of premature mortality associated with industrial PM 2.5 and O 3 is shown in Figs.5a and 5b.Eastern China and Northern India are hotspots for air pollution mortalities, driven by their large increases in surface PM 2.5 and O 3 concentrations and their large populations.East Asia (South Asia) account for 40 % (28 %) of the global all-cause mortalities associated with industrial PM 2.5 and 50 % (19 %) of the global respiratory mortalities associated with industrial O 3 .None of the other regions contribute over 10 % to the global mortalities associated with industrial air pollution.North America and Europe account for a smaller fraction of global industrial PM 2.5 mortality (2 % and 5 %, respectively) than of global industrial O 3 mortality (7 % and 8 %, respectively) while Africa has the reverse results (7 % vs. 5 %).

Attribution of premature mortalities associated with industrial air pollution
In Sect.4, changes in surface O 3 and PM 2.5 from preindustrial to present (2000) are attributed to changes in air pollutant emissions, climate and CH 4 concentrations.Here, we estimate the mortality responses using Eq. ( 1) with concentration changes driven by each factor separately.

Impact of changes in emissions of short-lived species
We estimate the global mortality response associated with industrial PM 2.5 and O 3 pollution resulting from changes in air pollutant emissions only ("2000"-"2000CL1860EM").We find that if air pollutant emissions in year 2000 had remained at 1860 levels, 1.48 (95 % CI, 1.01-1.93) million all-cause mortalities associated with PM 2.5 exposure and Introduction

Conclusions References
Tables Figures

Back Close
Full

Impact of historical climate change
We next estimate the global mortality response associated with industrial PM 2.5 and O 3 pollution resulting from climate change (and stratospheric O 3 depletion) ("2000" -"1860CL2000EM").We find that if climate and stratospheric O 3 in 2000 were the same as in 1860, about 86 (95 % CI, 58-114) thousand all-cause mortalities associated with PM 2.5 exposure and 7 (95 % CI, 2-12) thousand respiratory mortalities associated with O 3 exposure could have been avoided.

Impact of increased global methane
Finally, we evaluate the effects of CH 4 concentration increases on global premature mortalities associated with air pollution.To be entirely consistent with the previous comparisons, we would compare PM 2.5 and O 3 in "2000" with that in a simulation identical to "2000" except with the lower boundary condition of global CH 4 concentration specified at 1860 instead of 2000 levels in its tropospheric chemistry calculation.Unfortunately, such a simulation is not available.However, two of our available simulations ("1860CL2000EM" and "1860ALL2000EM") although simulating 1860 climate, differ only in their treatments of CH 4 in tropospheric chemistry.This allows us to estimate the effect of increased CH 4 on premature mortality due to O 3 and PM 2.

Comparison of factors contributing to premature mortality
We further compare the mortality response associated with each factor (emission, climate and CH 4 , see Figs. 5c-h) with total mortality associated with changes in industrial pollution regionally and globally to provide an approximation of the relative contribution of each factor to total air pollution mortalities in 2000 (the number of mortalities in each region due to each factor is summarized in Table S3 of the Supplement).For each region (i ), we obtain a Normalized Mortality Contribution (NMC) in the following way: Mortality responses driven by one anthropogenic factor Mortality responses driven by all anthropogenic factors (2) Figure 6 shows NMC for each region.Due to the non-linearity in the health impact function, chemistry, and chemistry-climate system, the value of each bar in Fig. 6 is close to, but not exactly 1.
Global premature mortality associated with industrial PM 2.5 , is dominated by increased emissions of reactive air pollutants (∼ 95 %), however, climate change is influential, contributing a global NMC of 5 %.Regionally, contributions of climate change to total premature mortality associated with industrial PM 2.5 can be as high as 12 % with the highest values over Europe and Australia.The impact of increased CH 4 concentrations on PM 2.5 and associated premature mortality globally is insignificant (Table 3).
The global premature respiratory mortality associated with industrial O 3 is also dominated by increased emissions of short-lived air pollutants (more than 85 %).However increases in CH 4 are also influential, contributing a global NMC of 13 % while the contribution of climate change is small with a global NMC of about 1 %.On a regional scale, NMC of increased CH 4 ranges from 10 to 33 % with the largest increases in excess mortalities in regions where increases in short-lived air pollutant emissions are relatively low (i.e., Australia, South America and Africa).In this study, we apply the GFDL Atmospheric Model version 3 (AM3), a chemistryclimate model, to examine changes in surface O 3 and PM 2.5 from the preindustrial period to the end of the 20th century and associated changes in premature mortality.Simulated global population-weighted PM 2.5 and H-O 3 (health-related O 3 , defined as the maximum 6-month mean of 1-h daily maximum O 3 in a year) concentrations increase by 8 ± 0.16 µg m −3 and 30 ± 0.16 ppbv, respectively, from the preindustrial period to present-day (1860 to 2000).We quantify excess mortalities attributable to industrial air pollution and find that around year 2000, industrial PM 2.5 and O 3 are associated with 1.5 (95 % CI, 1.0-2.0) million annual all-cause mortalities and 0.37 (95 % CI, 0.13-0.59) million annual respiratory mortalities, respectively.India and China suffer most from industrial air pollution mortality as they have experienced large increases in air pollution levels and have large exposed populations.We further evaluate the relative importance of changes in emissions of short-lived species, climate, and CH 4 concentrations in driving changes in PM 2.5 , O 3 and associated premature mortalities.
We find that increases in short-lived air pollutant emissions from 1860 to 2000 lead to 7.5 ± 0.19 µg m −3 and 25 ± 0.30 ppbv increases in PM 2.5 and H-O 3 concentrations respectively, accounting for a majority (94 % and 83 %) of total increases in these two species over this period.Changes in emissions of short-lived pollutants account for over 95 % of all-cause mortalities associated with industrial PM 2.5 and over 85 % of respiratory mortalities associated with industrial O 3 .
CH 4 concentration increase is the second most important driver of H-O 3 increases, causing an increase of 4.3 ± 0.33 ppbv and accounting for almost 15 % of the total increase of H-O 3 from 1860 to 2000.CH 4 contributes nearly 15 % of the total respiratory mortalities associated with industrial O 3 .CH 4 has negligible effects on PM 2.5 and has an insignificant effect on total all-cause mortalities associated with PM 2.5 .Introduction

Conclusions References
Tables Figures

Back Close
Full Changing climate has a small role in driving O 3 change during the industrial period, causing an increase of 0.5 ± 0.28 ppbv (2 % of the total increase of H-O 3 ) from 1860 to 2000.Climate change is the second most important driver of changes in PM 2.5 during the industrial period, causing an increase of 0.4 ± 0.17 µg m −3 (5 % of its total increase) from 1860 to 2000.The effect of climate change on industrial air pollution mortalities is small but non-negligible for both PM 2.5 and O 3 , accounting for less than 5 % and approximately 2 % of changes in mortality associated with these species during the industrial period, respectively.
The contribution together of climate change and CH 4 concentration increases to excess mortalities over various regions due to industrial air pollution ranges from 1 % to 13 % for all-cause mortality associated with industrial PM 2.5 and from 8 % to 33 % for respiratory mortality associated with O 3 .In terms of respiratory mortalities associated with industrial O 3 exposure, increased CH 4 concentrations alone contribute more than 20 % over South America, Europe, Africa, Middle East and Rest of Asia.Recent projections indicate that over Europe and the United States local O 3 precursor emissions are likely to continue to decrease after 2000 (van der A et al., 2008;Richter et al., 2005).In the mean time, CH 4 is projected to increase in almost all SRES and RCP emission scenarios (except RCP2.6 and SRES B2).As a result, the relative contribution of increased CH 4 to O 3 mortality will likely continue to rise, increasing the relative health benefits of CH 4 mitigation.
Increases in CH 4 concentrations since preindustrial time are highly influential on the oxidizing capacity of the atmosphere, as CH 4 increases are the most important factor driving changes in both tropospheric OH and H 2 O 2 abundance of all the factors discussed in this paper.Especially for OH, the primary atmospheric oxidant, historical increases in CH 4 drive its decreases, which will decrease its control of the buildup of reduced air pollutants such as CO and SO 2 .
As the benefit of CH 4 reduction does not depend on its location, for cleaner regions, such as Europe, South America and Australia (where we find mortality burdens to be more sensitive to CH 4 concentration than other regions), identifying the least cost Introduction

Conclusions References
Tables Figures

Back Close
Full      Full Discussion Paper | Discussion Paper | Discussion Paper | ) and atmospheric oxidants (i.e., OH, H 2 O 2 , O 3 ).Changes in O 3 Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | is the atmospheric component of the GFDL atmosphere-ocean coupled climate model CM3.AM3 is designed to address key emerging issues in climate science, including aerosol-cloud interactions and chemistryclimate feedbacks.It is GFDL's first global atmospheric model to include the indirect effects of cloud-aerosol interactions (with 16 interactive aerosol species) and of tropospheric and stratospheric chemistry (with 81 gas species) coupled with climate.Detailed chemistry, emissions, and deposition processes in AM3 are described in Naik et al. (2012) with additional details described below, and transport (advection, vertical diffusion and convection) along with physics are described by Donner et al. (2011).The model uses a finite-volume dynamical core with a 6 × 48 × 48 cubed-sphere horizontal Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | 2.5 observations are not available) for the 2000 simulation with observations over the United States (the U.S. Air Quality System, US-AQS), Europe (EMEP) and East Asia (Acid Deposition Monitoring Network in East Asia, EANET).Measured concentrations are averaged over corresponding model grids for 1997-2003 and 1995-2004 for the USAQS (PM 2.5 ) and EMEP (sulfate) datasets, respectively.Asian sulfate observations are collected from Liu et al. (2009) and Zhang Introduction Discussion Paper | Discussion Paper | Discussion Paper | et al.
5 and O 3 due to "industrial pollution" from preindustrial time (year 1860) to the present (year 2000) (TOTAL) in Sect.4.1.In the following subsections, we analyze the causes of these changes and attribute them to three factors: changes in emissions of air pollutants (EMIS, Sect.4.2), climate, (CLIM, Sect.4.3) and CH 4 concentrations (TCH4, Sect.4.4).As atmospheric OH and H 2 O 2 control the gas-and aqueous-phase production of major PM 2.5 components (such as sulfate) and are important for explaining PM 2.5 changes especially when discussing CH 4 effects, we also examine changes in their concentrations throughout this section.4.1 Total changes from 1860 to 2000 Distributions of annual mean surface PM 2.5 and health-relevant O 3 (defined as the maximum 6-month mean of 1-h daily maximum O 3 in a year, denoted as H-O 3 ) in year 2000 are shown in Figs.3a and 3b.Global population-weighted concentrations Introduction Discussion Paper | Discussion Paper | Discussion Paper | of PM 2.5 and H-O 3 are 13 ± 0.15 µg m −3 and 61 ± 0.20 ppbv ( 5 and O 3 from 1860 to 2000 are smaller than those reported in Anenberg et al. (2010) (15.0 µg m −3 and 37.1 ppbv), in large part due to the use of different present and preindustrial emissions in the two studies.Different from the emission inventory we use (Sect.2.2), the 2000 emissions of Anenberg et al. (2010) were obtained by scaling their standard 1990 emissions (taken from EDGAR v2.0) by the 2000 to 1990 ratio of SRES emissions (Nakicenovic Screen / Esc Printer-friendly Version Interactive Discussion Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | in Naik et al., 2012).According to Mieville et al. (2010), in boreal regions, burnt Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Factors contributing to increases over these regions include: (1) increased O 3 production due to higher water vapor leading to more abundant HO x ; (2) warmer global average temperature (+0.5 • ) from 1860 to 2000, increases photochemistry rates and decreases net formation of peroxyacetyl nitrate (PAN, CH 3 C(O)OONO 2 ), a reservoir species for NO x , leaving more NO x available over source regions and promoting local O 3 production; and Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | CH 4 increases (from 800 ppbv in 1860 to 1750 ppbv in 2000) result in an increase in the global population-weighted H-O 3 concentration of 4.3 ± 0.33 ppbv (Table 3, TCH4), accounting for almost 15 % of the total H-O 3 produced during the industrial period.Discussion Paper | Discussion Paper | Discussion Paper | 4 million respiratory mortalities would have been avoided in 2000 if surface PM 2.5 and O 3 had remained at 1860 levels (i.e., anthropogenic and biomass burning emissions of air pollution, CH 4 and climate had all remained the same in 2000 as they were in 1860).If we apply a low concentration threshold (LCT) of 5.8 µg m −3 and 33.3 ppbv (the lowest values in the ACS studies), premature mortalities associated with industrial PM 2.5 and O 3 are 15 % and 11 % lower, respectively.These relative differences are smaller here than Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | 0.33 (95 % CI, 113-522) million respiratory mortalities associated with O 3 exposure could have been avoided.
5 exposure We assume the bias due to the difference between 1860 and 2000 climates is small.Our results suggest that if CH 4 had remained at 1860 levels, about 50 (95 % CI, 17-82) thousand respiratory mortalities would have been avoided due to lower O 3 concentrations resulting from less CH 4 Discussion Paper | Discussion Paper | Discussion Paper | Thus respiratory mortalities from industrial O 3 over relatively clean regions are more affected by rising global background CH 4 concentrations than other regions.
Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | level of Ozone Depleting Substances is set to pre-1950 levels.b Simulations run for the Atmospheric Chemistry and Climate Model Inter-comparison Project (ACCMIP)Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper |
Table 3).Both PM 2.5 and H-O 3 show maximum concentrations close to source regions in East Asia, Eastern United States, India and Central Africa.High concentrations of PM 2.5 (ranging from 20 to 40 µg m −3 ) over Northern Africa and the Middle East are associated with strong Anenberg et al. (2010)are all lower than in this study (see Supplement, tableS2).Differences in 2000 emissions are generally larger than differences in the 1860 emissions of the two inventories.The resulting differences in emissions between the two studies are large.For example, the historical increase in SO 2 emissions in this study and inAnenberg et al. (2010)is 100 and 150 Tg per year, respectively; BC emission changes differ by even more, 4.4 vs.10 Tg C per year, respectively; differences in NO x emission changes are 31.7 vs. 35.0TgNper year.The greater emission changes from 1860 to 2000 inAnenberg et al. (2010)than in our study drive their higher estimate of changes in surface PM 2.5 and O 3 .We chose our emission inventory because it is newly developed to capture the best current information.It is widely used in various climate (and Anenberg et al. (2010)(Lamarque et al., , 2010))C AR5 report and the Atmospheric Chemistry and Climate Model Inter-comparison Project (ACCMIP)(Lamarque et al., 2012(Lamarque et al., , 2010)).Finally, in addition to differences in emission inventories, our study intrinsically differs fromAnenberg et al. (2010)as changes in PM 2.5 and O 3 from 1860-2000 here result from changes in climate, ODS and CH 4 as well as from emission changes.

Table 1 .
Model simulation configurations.All simulations are run for 11 yr with the first year used for spin-up (SST: sea surface temperature; SIC: sea ice; WMGG: well mixed greenhouse gases; ODS: ozone depleting substances).

Table 3 .
Year 2000 global population-weighted annual mean surface PM 2.5 and H-O 3 (healthrelevant O 3 , defined in Sect.2), tropospheric mean mass-weighted OH and H 2 O 2 concentrations, and their changes from 1860 to 2000 driven by all changes (TOTAL), changes in emissions of short-lived reactive pollutants (EMIS), changes in climate (CLIM), and changes in tropopsheric CH 4 concentration(TCH 4 ).Results are reported for 10-yr model simulations as annual average ± standard deviation.

Table 4 .
Premature mortalities in 2000 associated with industrial air pollution.Values are calculated as in Eq. (1), using ACS health impact functions, concentration difference in annual PM 2.5 and H-O 3 between "1860" and 2000" simulations, WHO baseline mortality rate and population in the year 2000.The 95 % confidence intervals are shown in brackets.