Impacts of aviation fuel sulfur content on climate and human health

18 Aviation emissions impact both air quality and climate. Using a coupled tropospheric chemistry-aerosol 19 microphysics model we investigate the effects of varying aviation fuel sulfur content (FSC) on premature 20 mortality from long-term exposure to aviation-sourced PM2.5 (particulate matter with a dry diameter of <2.5 21 μm) and on the global radiation budget due to changes in aerosol and tropospheric ozone. We estimate that 22 present-day non-CO2 aviation emissions with a typical FSC of 600 ppm result in ~3,600 [95% CI: 1,310–5,890] 23 annual premature mortalities globally due to increases in cases of cardiopulmonary disease and lung cancer, 24 resulting from increased surface PM2.5 concentrations. We quantify the global annual mean combined 25 radiative effect (REcomb) of non-CO2 aviation emissions as –13.3 mW m; from increases in aerosols (direct 26 radiative effect and cloud albedo effect) and tropospheric ozone. 27 28 Ultra-low sulfur jet fuel (ULSJ; FSC = 15 ppm) has been proposed as an option to reduce the adverse health 29 impacts of aviation-induced PM2.5. We calculate that swapping the global aviation fleet to ULSJ fuel would 30 reduce the global aviation-induced mortality rate by ~620 [95% CI: 230–1020] mortalities a and increase 31 REcomb by +7.0 mW m. 32 33 We explore the impact of varying aviation FSC between 0–6000 ppm. Increasing FSC increases aviation34 induced mortality, while enhancing climate cooling through increasing the aerosol cloud albedo effect (CAE). 35 We explore the relationship between the injection altitude of aviation emissions and the resulting climate and 36 air quality impacts. Compared to the standard aviation emissions distribution, releasing aviation emissions at 37 the ground increases global aviation-induced mortality and produces a net warming effect, primarily through 38 a reduced CAE. Aviation emissions injected at the surface are 5 times less effective at forming cloud 39 condensation nuclei, reducing the aviation-induced CAE by a factor of 10. Applying high FSCs at aviation cruise 40 altitudes combined with ULSJ fuel at lower altitudes results in reduced aviation-induced mortality and 41 increased negative RE compared to the baseline aviation scenario. 42 43 44 45 46 47 48 49 50 51 52

Abstract.Aviation emissions impact both air quality and climate.Using a coupled tropospheric chemistry-aerosol microphysics model we investigate the effects of varying aviation fuel sulfur content (FSC) on premature mortality from longterm exposure to aviation-sourced PM 2.5 (particulate matter with a dry diameter of < 2.5 µm) and on the global radiation budget due to changes in aerosol and tropospheric ozone.We estimate that present-day non-CO 2 aviation emissions with a typical FSC of 600 ppm result in ∼ 3600 [95 % CI: 1310-5890] annual premature mortalities globally due to increases in cases of cardiopulmonary disease and lung cancer, resulting from increased surface PM 2.5 concentrations.We quantify the global annual mean combined radiative effect (RE comb ) of non-CO 2 aviation emissions as −13.3 mW m −2 ; from increases in aerosols (direct radiative effect and cloud albedo effect) and tropospheric ozone.
Ultra-low sulfur jet fuel (ULSJ; FSC = 15 ppm) has been proposed as an option to reduce the adverse health impacts of aviation-induced PM 2.5 .We calculate that swapping the global aviation fleet to ULSJ fuel would reduce the global aviation-induced mortality rate by ∼ 620 [95 % CI: 230-1020] mortalities a −1 and increase RE comb by +7.0 mW m −2 .
We explore the impact of varying aviation FSC between 0 and 6000 ppm.Increasing FSC increases aviation-induced mortality, while enhancing climate cooling through increasing the aerosol cloud albedo effect (CAE).We explore the relationship between the injection altitude of aviation emissions and the resulting climate and air quality impacts.Compared to the standard aviation emissions distribution, releasing aviation emissions at the ground increases global aviation-induced mortality and produces a net warming effect, primarily through a reduced CAE.Aviation emissions injected at the surface are 5 times less effective at forming cloud condensation nuclei, reducing the aviation-induced CAE by a factor of 10.Applying high FSCs at aviation cruise altitudes combined with ULSJ fuel at lower altitudes results in reduced aviation-induced mortality and increased negative RE compared to the baseline aviation scenario.
Aviation emits a range of gas-phase and aerosol pollutants that can influence climate.Emissions of carbon dioxide (CO 2 ) from aviation warm the climate (Lee et al., 2009(Lee et al., , 2010)).Emissions of nitrogen oxides (NO x ) warm the climate through tropospheric ozone (O 3 ) formation, which acts as a greenhouse gas, and cool climate via a decrease in the lifetime of the well-mixed greenhouse gas methane (CH 4 ) through increases in the OH radical (Holmes et al., 2011;Myhre et al., 2011).Sulfate and nitrate aerosols, formed from aviation sulfur dioxide (SO 2 ) and NO x emissions and through altered atmospheric oxidants, lead to a cooling (Unger, 2011;Righi et al., 2013;Dessens et al., 2014), and black carbon (BC) emissions result in a warming (Balkanski et al., 2010).Additionally, the formation of persistent linear contrails and contrail-cirrus from aircraft leads to warming (Lee et al., 2010;Rap et al., 2010;Burkhardt and Karcher, 2011).Overall, aviation emissions are thought to have a warming impact on climate, with net radiative forcing (RF) estimated as +55 mW m −2 (excluding cirrus cloud enhancement) (Lee et al., 2010).
Aviation emissions can increase atmospheric concentrations of fine particulate matter with a dry diameter of < 2.5 µm (PM 2.5 ).Short-term exposure to PM 2.5 can exacerbate existing respiratory and cardiovascular ailments, while long-term exposure can result in chronic respiratory and cardiovascular diseases, lung cancer, chronic changes in physiological functions and mortality (Pope et al., 2002;World Health Organisation, 2003;Ostro, 2004).In the US aviation emissions are estimated to lead to adverse health effects in ∼ 11 000 people (ranging from mortality, respiratory ailments and hospital admissions due to exacerbated respiratory conditions) and ∼ 23 000 work loss days per annum (Ratliff et al., 2009).Landing and take-off aviation emissions increase PM 2.5 concentrations, particularly around airports (Woody et al., 2011), increasing US mortality rates by ∼ 160 per annum.
Previous studies have estimated the number of premature mortalities due to exposure to pollution resulting from aviation emissions.Barrett et al. (2012Barrett et al. ( , 2010) ) used the methodology of Ostro (2004) to estimate that aviation emissions are responsible for ∼ 10 000 premature mortalities a −1 due to increases in cases of cardiopulmonary disease and lung cancer.Yim et al. (2015) using the same methodology but with the inclusion of the Rapid Dispersion Code (RDC) to simulate the local air quality impacts of aircraft ground level emissions estimated 13 920 (95 % CI: 7220-20 880) mortalities a −1 .Morita et al. (2014), using the integrated exposure-response (IER) model from Burnett et al. (2014) to derive relative risk (RR), estimate that aviation results in 405 (95 % CI: 182-648) mortalities a −1 due to increases in cases of lung cancer, stroke, ischemic heart disease, trachea, bronchus, and chronic obstructive pulmonary disease.Jacobson et al. (2013) estimate 310 (95 % CI: −400 to 4300) mortalities a −1 from aviation emissions due to cardiovascular effects.Taking these studies in account, the different methodologies applied and modes of mortality investigated aviation is estimated to be responsible for between 310 and 13 920 mortalities a −1 .
The introduction of cleaner fuels and pollution control technologies can improve ambient air quality and reduce adverse health effects of fossil fuel combustion (World Health Organisation, 2005).One proposed solution to reduce the adverse health effects of aviation-induced PM 2.5 is the use of ultra-low sulfur jet fuel (ULSJ), reducing the formation of sulfate aerosol (Barrett et al., 2012(Barrett et al., , 2010;;Ratliff et al., 2009;Hileman and Stratton, 2014).ULSJ fuels typically have a fuel sulfur content (FSC) of 15 ppm, compared with an FSC of between 550 and 750 ppm in standard aviation fuels (Barrett et al., 2012).The current global regulatory standard for aviation fuel is a maximum FSC of 3000 ppm (Ministry of Defence, 2011;ASTM International, 2012).
Despite the potential for decreased emission of SO 2 , application of ULSJ fuel will not completely remove the impacts of aviation on PM 2.5 .It is estimated that over a half of aviation-attributable surface-level sulfate is associated with oxidation of non-aviation SO 2 by OH produced from aviation NO x emissions, and not directly produced from aviationemitted SO 2 (Barrett et al., 2010).Therefore, even a completely desulfurized global aviation fleet would likely contribute a net source of sulfate PM 2.5 .Nevertheless, previous work has shown that the use of ULSJ fuel reduces global aviation-induced PM 2.5 by ∼ 23 %, annually avoid-ing ∼ 2300 (95 % CI: 890-4200) mortalities (Barrett et al., 2012).
Altering the sulfur content of aviation fuel also modifies the net climate impact of aviation emissions.A reduction in fuel sulfur content reduces the formation of cooling sulfate aerosols (Unger, 2011;Barrett et al., 2012), increasing the net warming effect of aviation emissions.The roles of sulfate both in climate cooling and in increasing surface PM 2.5 concentrations mean that policy makers must consider both health and climate when considering effects from potential reductions in sulfur emissions from a given emissions sector (Fiore et al., 2012).
In this study, we investigate the impacts of changes in the sulfur content of aviation fuel on climate and human health.A coupled tropospheric chemistry-aerosol microphysics model is used to quantify global atmospheric responses in aerosol and O 3 to varying FSC scenarios.Radiative effects due to changes in tropospheric O 3 and aerosols are calculated using a radiative transfer model, while the impacts of changes in surface PM 2.5 on human health are estimated using concentration response functions.Using a coupled tropospheric chemistry-aerosol microphysics model that includes nitrate aerosol allows us to assess the impacts of nitrate and aerosol indirect effects in addition to the ozone and aerosol direct effects that have been more routinely calculated.

Model description
We use GLOMAP-mode (Mann et al., 2010), embedded within the 3-D off-line Eulerian chemical transport model TOMCAT (Arnold et al., 2005;Chipperfield, 2006).Meteorology (wind, temperature and humidity) and largescale transport is specified from interpolation of 6-hourly European Centre for Medium Range Weather Forecasts (ECMWF) reanalysis (ERA-40) fields (Chipperfield, 2006;Mann et al., 2010).Cloud fraction and cloud top pressure fields are taken from the International Satellite Cloud Climatology Project (ISCCP-D2) archive for the year 2000 (Rossow and Schiffer, 1999).
GLOMAP-mode is a two-moment aerosol microphysics scheme representing particles as an external mixture of seven size modes (four soluble and three insoluble) (Mann et al., 2010).We use the nitrate-extended version of GLOMAPmode (Benduhn et al., 2016) which, as well as tracking sizeresolved sulfate, BC, OC, sea-salt and dust components, also includes a dissolution solver to accurately characterise the size-resolved partitioning of ammonia and nitric acid into ammonium and nitrate components in each soluble mode.Aerosol components are assumed to be internally mixed within each mode.GLOMAP-mode includes representations of nucleation, particle growth via coagulation, condensation and cloud processing, wet and dry deposition, and in-and below-cloud scavenging (Mann et al., 2010).
TOMCAT includes a tropospheric gas-phase chemistry scheme (inclusive of O x -NO y -HO x ), treating the degradation of C 1 -C 3 non-methane hydrocarbons (NMHCs) and isoprene, together with a sulfur chemistry scheme (Spracklen et al., 2005;Breider et al., 2010;Mann et al., 2010).The tropospheric chemistry is coupled to aerosol as described in Breider et al. (2010).
The nitrate-extended version of the TOMCAT-GLOMAPmode coupled model used in this investigation employs a hybrid solver to simulate the dissolution of semi-volatile inorganic gases (such as H 2 O, HNO 3 , HCl and NH 3 ) into the aerosol-liquid-phase.
Emissions of DMS are calculated using monthly mean sea-water concentrations of DMS from (Kettle and Andreae, 2000), driven by ECMWF winds and sea-air exchange parameterisations from Nightingale et al. (2000).Emissions of SO 2 are included from both continuous (Andres and Kasgnoc, 1998) and explosive volcanoes (Halmer et al., 2002), and wildfires for year 2000 (Van Der Werf et al., 2003;Dentener et al., 2006).Anthropogenic SO 2 emissions (including industrial, power plant, road transport, off-road transport and shipping sectors) are representative of the year 2000 (Cofala et al., 2005).Emissions of monoterpenes and isoprene are from Guenther et al. (1995).NH 3 emissions are from the EDGAR inventory (Bouwman et al., 1997).NO x emissions are considered from anthropogenic (Lamarque et al., 2010), natural (Lamarque et al., 2005) and biomass burning (van der Werf et al., 2010) sources.
Annual mean emissions of BC and OC aerosol from fossil fuel and biofuel combustion are from Bond et al. (2004).Monthly wildfire emissions are taken from the GFED v1 (Global Fire Emissions Database) for the year 2000 (Van Der Werf et al., 2003).For primary aerosol emissions we use geometric mean diameters (D g ) with standard deviations as described by Mann et al. (2010).
Here, we ran simulations at a horizontal resolution of 2.8 • × 2.8 • with 31 hybrid σ -p levels extending from the surface to 10 hPa.All simulations were conducted for 16 months from September 1999 to December 2000 inclusive, with the first 4 months discarded as spin-up time.

Model evaluation
GLOMAP has been extensively evaluated against observations including comparisons of speciated aerosol mass (Mann et al., 2010;Spracklen et al., 2011b), aerosol number (Mann et al., 2010;Spracklen et al., 2010) and cloud condensation nuclei (CCN) concentrations (Spracklen et al., 2011a).TOMCAT simulated fields have been evaluated against observations, with CO and O 3 evaluated against aircraft observations (Arnold et al., 2005) ozone against satellite observations (Richards et al., 2013), along with O 3 evaluated against satellite observations (Chipperfield et al., 2015).Benduhn et al. (2016) shows that simulated surface concentrations of NO 3 and NH 4 are in reasonable agreement with observations in Europe, the US and East Asia.Here we focus our evaluation on the aerosol vertical profile as well as nitrate aerosol which has not been evaluated previously.
Figure 1 presents simulated sulfate, nitrate, ammonium and organic aerosol mass concentrations in comparison to airborne observations compiled by Heald et al. (2011).The supplementary information presents the flight paths of each of the aircraft field campaigns used in the study compiled by Heald et al. (2011) (Fig. S1 in the Supplement), and details of each of the aircraft field campaigns used (Table S1 in the Supplement).Observations were predominantly made using an Aerodyne aerosol mass spectrometer (AMS).Simulated profiles are for year 2000, while observational aerosol profiles are from field campaigns conducted between 2001 and 2008.
Overall we find the model overestimates sulfates [NMB = +16.9%], while underestimating nitrates [NMB = −60.7 %], ammonium [NMB = −47.1 %] and organic aerosols (OA) [NMB = −56.2%].Model skill varies depending on the conditions affecting each field campaign.To explore this, we use the broad stratification of the field campaigns into anthropogenic pollution, biomass burning and remote conditions as used by Heald et al. (2011)  The overestimation of sulfate aerosol is likely due to the decline in anthropogenic SO 2 emissions in Europe and the US between 2000 and 2008 (Vestreng et al., 2007;Hand et al., 2012).An underestimation of OA has been reported previously (Heald et al., 2011;Spracklen et al., 2011b) and is likely due to an underestimate in SOA formation in the model.Whitburn et al. (2015) found biomass burning emissions of NH 3 may be underestimated which would affect a number of our comparisons.
The model underestimation of organic and inorganic aerosol components in biomass burning influenced regions could partly be due to very concentrated plumes in these regions affecting campaign mean concentrations.There is a large uncertainty in biomass burning emissions and some evidence that they may be underestimated (Kaiser et al., 2012), which may contribute to the model bias.Biomass burning emissions also have large interannual variability (van der Werf et al., 2010;Wiedinmyer et al., 2011), meaning that using year-specific emissions might improve comparison against observations in these regions.Underestimation in Arctic inorganic aerosol, which will affect the ARCTAS comparisons, is a well-known problem in models, likely related to problems with model wet deposition and emissions (Shindell et al., 2008;Eckhardt et al., 2015).The model underestimate over West Africa (AMMA, DADEX and DODO campaigns) is likely due to a combination of errors in biomass burning emissions and poorly constrained emission sources from anthropogenic activity (Knippertz et al., 2015).
Regional model-observation comparison profiles presented in Fig. 2 demonstrate good agreement between the model and ozonesonde profiles, while demonstrating regional variations driven by variations in tropopause height, showing no evidence of systematic model bias in the upper troposphere.Notable differences are seen between simulated and observed ozone profiles over the Praha launch site in Western Europe, with the model greatly overestimating observed ozone.
Differences between model and observational profiles can in part be explained by the differences in years of simulation and observation, a poor representation of deep convection resulting in model underestimations in the tropics and overestimations downwind (Thompson et al., 1997), in tandem with reductions in anthropogenic NO x emissions over this time period (Konovalov et al., 2008).

Aviation emissions
Aircraft emit NO x , carbon monoxide (CO), SO 2 , BC, OC and hydrocarbons (HCs).The historical emissions data set for the CMIP5 (5th Coupled Model Intercomparison Project) model simulations used by the IPCC 5th Assessment Report only included NO x and BC aviation emissions (Lamarque et al., 2009).Recently there have been efforts to add HCs, CO and SO 2 emissions to aviation emission inventories (Eyers et  Here we develop a new 3-D civil aviation emissions data set for the year 2000, based on CMIP5 historical aviation emissions (Lamarque et al., 2009).The new data set includes emissions of NO x , CO, SO 2 , BC, OC, and HCs.In contrast to existing data sets which provide a general emissions index for HCs (Eyers et al., 2004) we speciate HCs as formaldehyde (HCHO), ethane (C 2 H 6 ), propane (C 3 H 8 ), methanol (CH 3 OH), acetaldehyde (CH 3 CHO), and acetone ((CH 3 ) 2 CO).
Table 1 describes our new emissions data set.NO x and BC emissions are taken directly from Lamarque et al. (2009).We calculate fuel burn from BC emissions data and the BC emissions index (Eyers et al., 2004) as used by Lamarque et al. (2009).Following DuBois and Paynter (2006), we assume that BC emissions scale linearly with fuel consumption.We estimate emissions for other species using our calculated aviation fuel burn in combination with published speciesspecific emissions indices (EI reported in g kg −1 of fuel).Emission indices for CO and SO 2 are from the FAA's aviation environmental design tool (AEDT) (Wilkerson et al., 2010).OC emissions are calculated using a BC : OC ratio of 4 (Bond et al., 2004); resulting in an EI within the range determined by Wayson et al. (2009).Speciated hydrocarbon emissions are calculated from experimental data following the methodology of Wilkerson et al. (2010) using experimental data from Knighton et al. (2007) and Anderson et al. (2006).
Our global aviation emissions typically lie within the range of previous studies (Table 1).Our SO 2 emissions are greater than those used by Wilkerson et al. (2010) for 2006, despite the use of the same EI.This is due to the greater global fuel burn considered by the base inventory used to develop our emissions inventory (Eyers et al., 2004;Lamarque et al., 2010).Our estimated OC emissions are lower than the emissions estimated in the AEDT 2006 inventory, due to the lower EI applied here.The lower EI OC applied here (in comparison to Wilkerson et al., 2010) is a due to the phase of flight considered when deriving the AEDT emissions inventory; where they derive EI OC focusing on airport operations at ground-level condition acknowledging the risk of overestimating aviation OC emissions, while in comparison we consider aircraft operations after ground idle conditions which risks underestimating aviation OC emissions.
We calculate the geometric mean diameter (D g ) for internally mixed BC / OC particles as 50.5 nm from the mean particle mass derived using the particle number emissions index (Eyers et al., 2004) and a constant standard deviation set to σ = 1.59 nm.

Fuel sulfur content simulations
To explore the impact of aviation FSC on climate and air quality we performed a series of 11 global model experiments (Table 2).In 7 of these model experiments FSC values were varied globally between zero and 6000 ppm.Three further simulations varied the vertical distribution of aviation emissions.The first simulation collapses all aviation emissions to ground level (GROUND), in order to compare an equivalent ground emission source and its effects.Two simulations (SWITCH1 and SWITCH2) use a low FSC (15 ppm) applied below the cruise phase of flight (< 8.54 km altitude) (Lee et al., 2009;Köhler et al., 2013) combined with a high FSC at altitudes above cruise level.The SWITCH1 scenario increases FSC in line with our HIGH scenario above 8.54 km, while in the SWITCH2 scenario, emissions are scaled such that total global sulfur emissions are the same as the standard simulation (NORM), resulting in a FSC of 1420 ppm above 8.54 km.Results from all simulations are compared against a simulation with aviation emissions excluded (NOAVI).

Radiative impacts
We calculate the aerosol direct radiative effect (aDRE), aerosol cloud albedo effect (aCAE) and tropospheric O 3 direct radiative effect (O3DRE) using the offline Edwards and Slingo (1996) radiative transfer model.The radiative transfer model considers six bands in the shortwave (SW) and nine bands in the longwave (LW), adopting a delta-Eddington 2 stream scattering solver at all wavelengths.The top-of-the-atmosphere (TOA) aerosol aDRE and aCAE are calculated using the methodology described in Rap et al. (2013) and Spracklen et al. (2011a), with the method for O3DRE as in Richards et al. (2013)  (CDNCs) using the monthly mean aerosol size distribution simulated by GLOMAP combined with parameterisations from Nenes and Seinfeld (2003), updated by Fountoukis and Nenes (2005) and Barahona et al. (2010).CDNC were calculated with a prescribed updraft velocity of 0.15 m s −1 over ocean and 0.3 m s −1 over land.Changes to CDNC were then used to perturb the effective radii of cloud droplets in lowand mid-level clouds (up to 600 hPa).The aDRE, aCAE and O3DREs for each aviation emissions scenario are calculated as the difference in TOA net (SW + LW) radiative flux compared to the NOAVI simulation.

Health effects
We calculate excess premature mortality from cardiopulmonary diseases and increases in cases of lung cancer due to long-term exposure to aviation-induced PM 2.5 (Ostro, 2004).
Using this function allows us to compare our premature mortality estimates with those from previous studies (Barrett et al., 2012;Yim et al., 2015) using the same concentration function; in future work estimates are required with updated methodologies (Burnett et al., 2014).PM 2.5 is used as a measure of likely health impacts because chronic exposure is associated with adverse human health impacts including morbidity and mortality (Dockery et al., 1993;Pope and Dockery, 2006).
We relate annual excess mortality to annual mean surface PM 2.5 via a concentration response function (CRF) (Ostro, 2004).This response function considers concentrations of PM 2.5 for a perturbed case (X) (defined by aviation emissions scenarios from Table 2) in relation to a baseline case with no aviation emissions (X 0 ) (NOAVI).To calculate excess mortality, the relative risk (RR) for both cardiopulmonary disease and lung cancer is calculated according to Ostro (2004) using a function of baseline (X 0 ) and perturbed (X) PM 2.5 concentrations, and the disease-specific, causespecific coefficient (β): β coefficients for cardiopulmonary disease mortality of 0.15515 [95 % CI = 0.05624 − 0.2541] and lung cancer of 0.232 [95 % CI = 0.086 − 0.379] are used (Pope et al., 2002;Ostro, 2004).The 95% confidence interval (CI) in β allow low-, mid-and high-range mortality values to be calculated.The attribution factor (AF) from the exposure to air pollution is calculated using Eq. ( 2): Excess mortality (E) for both cardiopulmonary disease and lung cancer is calculated using baseline mortality rates (B), the fraction of the population over 30 years old (P 30 ), along with the AF: Global population data are taken from the Gridded World Population (GWP; version3) project (Center for International Earth Science Information Network, 2012) with countryspecific data on the fraction of the population under 30.

Surface PM 2.5
Figure 3 shows the simulated impact of aviation emissions with standard FSC (FSC = 600 ppm; NORM) on surface PM 2.5 concentrations.Aviation increases annual mean PM 2.5 concentrations by up to ∼ 80 ng m −3 (relative to the NOAVI simulation) over Central Europe and Eastern China (Fig. 3a).Aviation emissions result in largest fractional changes in annual mean PM 2.5 concentrations (up to 0.8 %) over North America and Europe (Fig. 3b).
The use of ULSJ fuel (FSC = 15 ppm) reduces global annual mean surface aviation-induced PM North America [ sulfate = −2.9ng m −3 ; nitrate = +0.02ng m −3 ; ammonium = −0.5 ng m −3 ; and others = −0.01ng m −3 ] (Fig. 4b, c).Over North America, swapping to ULSJ fuel reduces aviation-induced PM 2.5 by 53.4 %, while a smaller reduction of 20.5 % is simulated over Europe.The smaller fractional change in PM 2.5 over Europe is caused by smaller reductions in aviation-induced sulfate [−55.9 %] and ammonium [−18.4 %] compared to over North America, which sees a reduction in ammonium of 41.6 % and a reduction in sulfates of 103 % indicating that over the US the ULSJ fuel scenario sees a reduction in sulfates in relation to a NOAVI scenario.
Complete desulfurization of jet fuel (FSC = 0 ppm; DESUL) reduces global mean aviation-induced surface PM 2.5 concentrations by 36.5 % [−1.43 ng m −3 ], with changes in sulfates [−1.40 ng m −3 ; −63.5 %] and ammonium [−0.24 ng m −3 ; −38.8 %] dominating.Under this scenario the reductions in surface sulfate PM 2.5 from aviation are 57.3 % over Europe and 105 % over North America.ULSJ fuel therefore gives similar results to complete desul-  furization, due to the very small sulfur emission from ULSJ fuel (Table 2).In summary, increases in FSC result in increased surface PM 2.5 , due to increased sulfate outweighing the small reductions in nitrate.Simulated changes in sulfate, nitrate, ammonium and total PM 2.5 are linear (R 2 > 0.99, p value < 0.001 globally and for all individual regions) with respect to FSC (Fig. 4).Larger emission perturbations would likely lead to a non-linear response in atmospheric aerosol.The impact of variations in FSC on PM 2.5 are regionally variable; over Europe changes in PM 2.5 concentrations are observed to be more sensitive to changes in FSC than over North America, and the global domain.
Figure 5 shows the impact of changing to ULSJ fuel on zonal mean sulfate and nitrate concentrations relative to standard fuel (NORM).Table 3 reports the global aerosol burden from aviation under different emission scenarios.With standard FSC (FSC = 600 ppm), the global aviation-induced aerosol burden is 16.9 Gg, dominated by sulfates (76.3 %) and nitrates (33.4 %).The use of ULSJ (FSC = 15 ppm) reduces the global aerosol burden from aviation by 26.8 %.Complete desulfurization of aviation fuel reduces the global aerosol burden from aviation by 28.4 %, with the global sulfate burden from aviation reduced by 71.6 % (Table 3).When aviation emissions contain no sulfur, aviation-induced sulfate is formed through aviation NO x -induced increases in OH concentrations, resulting in the oxidation of SO 2 from nonaviation sources (Unger et al., 2006;Barrett et al., 2010).
In line with previous work, we find that a substantial fraction of aviation sulfate can be attributed to aviation NO x emissions and not directly to aviation SO 2 emissions.We estimate that 36 % aviation-attributable sulfates formed at the surface are associated with aviation NO x emissions, compared to ∼ 63 % estimated by Barrett et al. (2010) using the GEOS-Chem model (both estimates for FSC = 600 ppm).Differences between model estimates can be attributed to differences in model chemistry and microphysics, and different aviation NO x emissions.We find desulfurization increases the aviation nitrate burden by 5.1 % (Table 3); although much of this increase occurs at altitudes well above the surface (Fig. 5) and so is not reflected in surface PM 2.5 concentrations.
We explored the impacts of NO x emission reductions in combination with fuel desulfurization.A scenario with desulfurized fuel and zero NO x emissions reduces the global aviation-induced aerosol burden by 88.3 % (Table 3), in comparison to a desulfurized only case (DESUL), where the aviation-induced aerosol burden is reduced by 28.4 %.Removal of aviation NO x and SO 2 emissions results in a 95.0 % reduction in aviation-induced global mean surface level aviation-induced PM 2.5 .These results imply that only limited sulfate reductions can be achieved through reducing FSC alone, with further reductions in aviation-induced PM 2.5 sulfates requiring additional controls on aviation NO x emissions.

Premature mortality
Figure 6 shows estimated annual premature mortalities (from cardiopulmonary disease and lung cancer) due to aviationinduced changes in PM 2.5 as a function of FSC.We estimate that aviation emissions with standard FSC (FSC = 600 ppm) cause 3600 [95 % CI: 1310-5890] premature mortalities each year, with 3210 [95 % CI: 1160-5250] mortalities a −1 due to increases in cases of cardiopulmonary disease and 390 [95 % CI: 150-640] mortalities a −1 due to increases in cases of lung cancer.Low-, mid-and high-range cause-specific coefficients (β) are used to account for uncertainty in the health impacts caused by exposure to PM 2.5 (Sect.2.5) (Ostro, 2004).Our estimated global mortality due to aviation emissions is greatest in the Northern Hemisphere, which accounts for 98.7 % of global mortalities.Europe and North America account for 42.3 and 8.4 % of mortality due to aviation emissions respectively.
Our estimate of the premature mortality due to aviation lies within the range of previous estimates (310-13 920 mortalities a −1 ) (Barrett et al., 2010(Barrett et al., , 2012;;Jacobson et al., 2013;Morita et al., 2014;Yim et al., 2004).Barrett et al. (2012) estimated ∼ 10 000 mortalities a −1 due to aviation, almost a factor of 3 higher than our central estimate.The greater aviation-induced mortality simulated by Barrett et al. (2012) can be attributed to greater aviation-induced surface PM 2.5 concentrations simulated in their study, particulary over highly populated areas.Their study simulated maximum aviation-induced PM 2.5 concentrations over Europe, eastern China and eastern North America greater than those in our simulations by factors of 5 for Europe and eastern China and 2.5 over eastern North America.Our aviation-induced sulfate concentrations compare well with Barrett et al. (2012), indicating that the resulting differences in aviation-induced surface PM 2.5 concentrations are a result of other aerosol components.Additionally, differences in mortality arise due to the use of different cause-specific coefficients (β) within the same CRF, as well as different population data sets.Morita et al. (2014) estimate that aviation is responsible for 405 [95 % CI: 182-648] mortalities a −1 .This lower estimate is primarily due to the mortality functions used, with Morita et al. (2014) using the integrated exposure response (IER) function as described by Burnett et al. (2014).The IER function considers a PM 2.5 concentration below which there is no perceived risk, reducing estimated impacts of aviation in regions of low PM 2.5 concentrations.
We estimate that aviation emissions with ULSJ fuel result in 2970 [95 % CI: 1080-4870] premature mortalities globally per annum.Therefore, changing from standard FSC to ULSJ would result in 620 [95 % CI: 230-1020] fewer premature mortalities globally per annum; a reduction in aviationinduced mortalities of 17.4 %.Regionally we find the implementation of an ULSJ fuel reduces annual mortality by 180 over Europe and by 110 over North America.Barrett et al. (2012) estimated that swapping to ULSJ fuel could result in ∼ 2300 [95 % CI: 890-4200] fewer premature mortalities globally per annum; a reduction of 23 %.In their work (using GEOS-Chem), the use of ULSJ reduces global mean PM 2.5 concentrations (sulfates, nitrates and ammonium) by 0.89 ng m −3 , less than the 1.61 ng m −3 reduction in PM 2.5 simulated here).Despite the greater reductions in global mean surface layer PM 2.5 concentrations simulated here, Barrett et al. (2012) simulate greater reductions in PM 2.5 over populated regions, resulting in greater reductions of aviation-induced mortality under the ULSJ scenario.Additionally, the GRUMPv1 population data set that Barrett et al. (2012) use resolves population data on a finer scale compared to the resolution of GPWv3 population data set used here (Center for International Earth Science Information Network, 2012); differences which could contribute to differences in estimates of mortality.We also estimate how aviation-induced mortality would change if FSC was increased.We find that increasing FSC to 3000 ppm (HIGH) would increase annual aviation-induced mortalities to 6030, an increase of 67.8 % in relation to standard aviation (NORM; FSC = 600 ppm).
Figure 8 shows the sensitivity of low level CCN concentrations to FSC.As with PM 2.5 , we find simulated changes in CCN are near linear with respect to FSC (R 2 > 0.99 and p value < 0.001 globally and for all individual regions).

Sensitivity of aerosol and ozone radiative effect
to FSC shows the calculated global mean net RE due to non-CO 2 aviation emissions.For standard FSC (FSC = 600 ppm) emissions the global mean combined RE is −13.3 mW m −2 .This combined radiative effect (RE comb ) results from a balance between a positive aDRE of +1.4 mW m −2 and O3DRE +8.9 mW m −2 , and a negative aCAE of −23.6 mW m −2 (Fig. 9).Our estimated aviation aerosol DRE [+1.4 mW m −2 ] lies in the middle of the range given by previous work.The aviation aerosol DRE has been previously assessed as highly uncertain, ranging between −28 and +20 mW m −2 (Righi et al., 2013) Our O3DRE estimate (+8.9 mW m −2 ), normalised by global aviation NO x emission to +10.5 mW m −2 Tg(N) −1 , is at the lower end of current estimates [7.4-37.0mW m −2 Tg(N) −1 ] ( Sausen et al., 2005;Köhler et al., 2008;Hoor et al., 2009;Lee et al., 2009;Holmes et , 2011;Myhre et al., 2011;Unger, 2011;Frömming et al., 2012;Skowron et al., 2013;Unger et al., 2013;Khodayari et al., 2014).This can be attributed to the lower net O 3 chemical production efficiency (OPE) within our model (1.33).Unger (2011)  We calculate that an aviation fleet utilising ULSJ fuel would result in a global annual mean RE comb of −6.3 mW m −2 [aDRE = +1.8mW m −2 ; aCAE = -16.8mW m −2 ; and O3DRE = +8.7 mW m −2 ].Thus, swapping from standard aviation fuel to ULSJ fuel reduces the net cooling effect from aviation-induced aerosol and O 3 by 7.0 mW m −2 , in comparison to the reduction of 3.3 mW m −2 estimated by Barrett et al. (2012).In our model, this change is primarily due a reduction in cooling from the aCAE of +6.7 mW m −2 combined with smaller contributions from an increased aDRE of +0.4 mW m −2 , and reduction in warming from the O3DRE of −0.12 mW m −2 (Fig. 9).
Increases in FSC result in reductions in the aerosol DRE (aDRE), changing from a positive aerosol DRE for low FSC scenarios, to a negative aerosol DRE for high FSC (FSC > 1200 ppm).As FSC is increased, we find the aCAE exhibits a larger cooling effect, i.e. becoming more negative with increases in FSC, increasing by a factor ∼ 5 as FSC is increased from 0 to 6000 ppm.The RE comb is dominated by these changes to the aCAE.Increases in FSC from 0 to 6000 ppm result in a greater negative (cooling) aviation- induced RE comb ; increasing in magnitude by a factor of ∼ 5 (−16.6 mW m −2 for FSC = 0 ppm to −82.1 mW m −2 for FSC = 6000 ppm) (Fig. 9).Therefore, we find that increases in FSC provide a cooling effect due to the dominating effect from aviation-induced aCAE.The different slopes in the relationship between estimated RE and mortality (Fig. 10) are driven by the range of coefficients used in the CRF.This highlights the considerable uncertainty in the health impacts caused by exposure to PM 2.5 .We note that uncertainty in the RE due to aerosol and ozone exists, but is not included in Fig. 9.
To assess how the vertical distributions of aviation SO 2 emissions influence human health and climate effects, we performed three additional simulations where we altered the vertical distribution of aviation SO 2 emissions (GROUND, SWITCH1 and SWITCH2 simulations).In these simulations the relationships between mortality and net RE deviate from the linear relationship seen when varying FSC between 0 and 6000 ppm (Fig. 10).
Releasing aviation emissions at the surface (GROUND case) increases global mean cloud level CCN by only 0.4 cm −3 relative to NOAVI; providing a reduction in CCN of 82.1 % [−1.89 cm −3 ] relative to the NORM case (i.e.GROUND-NORM).That is, injecting aviation emissions into the free troposphere in the standard scenario is over 5 times more efficient at increasing CCN concentrations compared to when the same emissions are released at the surface [GROUND CCN = 0.4 cm −3 ; NORM CCN = 2.3 cm −3 ]; both in relation to the NOAVI scenario.Similar behaviour has been demonstrated previously for volcanic SO 2 emissions by Schmidt et al. (2012), where volcanic SO 2 emissions injected into the free troposphere (FT) were more than twice as effective at producing new CCN compared to boundary layer emissions of DMS.Injection of aviation SO 2 emissions at the surface will increase both deposition rates and aqueous phase oxidation of SO 2 ; the latter resulting in the growth of existing CCN, but not the formation of new CCN.In contrast, when SO 2 is emitted into the FT the dominant oxidation mechanism is to H 2 SO 4 , leading to the formation of new CCN through particle formation and the condensational growth of particles to larger sizes.Subsequent entrainment of these new particles into the lower atmosphere results in enhanced CCN concentrations in low-level clouds.Reduced CCN formation when aviation emissions are injected at the surface has implications for the aCAE.When aviation emissions are released at the surface we calculate an aCAE of −2.3 mW m −2 ; a factor of 10 smaller than the standard aviation scenario.This demonstrates that low-level CCN concentrations and the aCAE are particularly sensitive to aviation emissions, because of the efficient formation of CCN when SO 2 emissions are injected into the FT.Injecting aviation emissions at the surface also results in an increase in the aDRE of +5.9 mW m −2 , resulting in an RE comb of +5.0 mW m −2 (Fig. 9).
Surface O 3 concentrations are also less sensitive to aviation when emissions are located at the surface.Global mean aviation-induced surface O 3 concentrations are reduced from 0.15 ppbv (NORM) to 0.03 ppbv when all emissions are in the surface layer.Releasing aviation emissions at the surface also reduces the global O 3 burden by 3.1 Tg.These perturbations in O 3 concentrations result in a reduction in the O 3 ra-diative effect from +8.9 mW m −2 (NORM; FSC = 600 ppm) to +1.5 mW m −2 (GROUND; FSC = 600 ppm) (Fig. 9).This is a reflection of increases in the OPE of NO x with increases in altitude due to lower background NO x and NMHC (nonmethane hydrocarbon) concentrations (Köhler et al., 2008;Stevenson and Derwent, 2009;Snijders and Melkers, 2011;Skowron et al., 2013).

Discussion and conclusions
We have used a coupled chemistry-aerosol microphysics model to estimate the impact of aviation emissions on aerosol and O 3 concentrations, premature mortality and radiative effect on climate.
We estimate that standard aviation (NORM; FSC = 600 ppm) is responsible for approximately 3600 premature mortalities annually due to increased surface layer PM 2.5 , in line with previous work (Barrett et al., 2012).We find that aviation-induced mortalities are highest over Europe, eastern North America and eastern China; reflecting larger regional perturbations in surface layer PM 2.5 concentrations.Comparing these estimates with total global premature mortalities from ambient air pollution from all anthropogenic sources (Lim et al., 2012), aviation is responsible for 0.1 % [0.04-0.18%] of annual premature mortalities.
We investigated the impact of varying aviation FSC over the range 0-6000 ppm.Increases in FSC lead to increases in surface PM 2.5 concentrations and subsequent increases in aviation-induced mortality.Increases in FSC also lead to a more negative RE comb due to enhanced aCAEs.We estimate that the use of ultra-low sulfur jet (ULSJ) fuel, with a FSC of 15 ppm, could prevent 620  mortalities annually compared to standard aviation emissions.Swapping to ULSJ fuel increases the global mean net RE by +7.0 mW m −2 compared to standard aviation emissions, largely due to a reduced aCAE.We calculate a larger warming effect from switching to ULSJ fuel than that assessed by Barrett et al. (2012), who did not evaluate changes in aCAE.
Absolute reductions in FSC result in limited reductions in aviation-induced surface layer PM 2.5 .We estimate that aviation-NO x emissions are responsible for 36.2 % of aviation-induced sulfate perturbations.Thus further reductions in aviation-induced PM 2.5 can potentially be achieved if NO x emission reductions are implemented in tandem with reductions to fuel sulfur content.
In line with previous work (Köhler et al., 2008;Stevenson and Derwent, 2009;Snijders and Melkers, 2011;Frömming et al., 2012;Skowron et al., 2013), decreasing the altitude at which O 3 -forming species are emitted results in a reduction in aviation-induced O 3 , and resulting O3DRE.This is due to the relationship between altitude and OPE, and the inverse relationship between altitude and background pollutant concentrations.We also explored the sensitivity of emission injection altitude on aerosol, mortality and aerosol RE.Injecting aviation emissions at the surface results in a reduction in global mean concentrations of PM 2.5 (relative to NORM), but with higher regional concentrations over central Europe and eastern America; resulting in higher annual mortalities due to aviation.We find that aviation emissions are a factor of 5 less efficient at creating CCN when released at the surface, resulting in an aCAE of −2.3 mW m −2 , a reduction of 90.1 % in relation to the standard aviation scenario.When aviation SO 2 emissions are injected into the free-troposphere, the dominant oxidation pathway is to H 2 SO 4 followed by particle formation and condensational growth of new particles to larger sizes.Subsequent entrainment of these new particles into the lower atmosphere leads to increased CCN concentrations and impacts on cloud albedo.Aviation SO 2 emissions are therefore particularly efficient at forming CCN with resulting impacts on cloud albedo.
We explored the impact of applying altitude-dependent variations in aviation FSC.We tested a scenario with high FSC in the free troposphere and low FSC near the surface, resulting in the same global aviation sulfur emission as the standard aviation scenario.In this scenario, aviation-induced premature mortalities were reduced by 2.4 % [−90 mortalities a −1 ] and the magnitude of the negative RE comb was increased by 36.6 %, providing an additional cooling impact of climate of −4.88 mW m −2 .
Our simulations suggest that the climate and air quality impacts of aviation are sensitive to FSC and the altitude of emissions.We explored a range of scenarios to maximise climate cooling and reduce air quality impacts.Use of ULSJ fuel (FSC = 15 ppm) at low altitude combined with high FSC in the free troposphere results in increased climate cooling whilst reducing aviation mortality.More complicated emission patterns, for example, use of high FSC only whilst over oceans might further enhance this effect.However, we note that the greatest reduction in aviation-induced mortality is simulated for complete desulfurization of aviation fuel.Given the uncertainty in both the climate and air quality impacts of aerosol and ozone, additional simulations from a range of atmospheric models are required to explore the robustness of our calculations.Finally, we note that our calculations are limited to calculation of aviation-induced RE.Future work needs to assess the complex climate impacts of altering aviation FSC.Future work needs to estimate the health impacts of aviation using newly available concentration response functions (Burnett et al., 2014).

Data availability
Ozone observations used in this work can be acquired from Tilmes et al. (2012).Aerosol data are described in Heald et al. (2011).CMIP5 aviation emissions are available from http://tntcat.iiasa.ac.at:8787/RcpDb/dsd?Action= htmlpage&page=about.Model outputs are available on request from the corresponding author.
The Supplement related to this article is available online at doi:10.5194/acp-16-10521-2016-supplement.

Figure 1 .
Figure 1.Comparison of observed (Obs) and simulated (Mod) (a) sulfate; (b) nitrate; (c) ammonium, and (d) organic aerosol mass concentrations.Observations are from airborne field campaigns compiled by Heald et al. (2011).Mean values are represented by black dots, median values as shown by horizontal lines, while boxes denote the 25th and 75th percentiles, and whiskers denote the 5th and 95th percentile values.

Figure 2 .
Figure 2. Comparison of observed (solid lines) and simulated (dashed lines) ozone profiles.Observations are taken from ozonesonde observations, and arranged by launch location regions according to Tilmes et al. (2012).
Figure4shows the impact of aviation emissions on global and regional mean PM 2.5 concentrations as a function of FSC.With standard FSC (FSC = 600 ppm), aviation increases global mean surface PM 2.5 concentrations by 3.9 ng m −3 ; with increases in PM 2.5 dominated by sulfates [56.2 %], nitrates [26.0 %] and ammonium [16.0 %].Aviation emissions increase European annual mean PM 2.5 concentrations by 20.3 ng m −3 (Fig.4b), substantially more than over North America (Fig.4c) where an annual mean increase of 6.3 ng m −3 is simulated.Increased PM 2.5 is dominated by nitrates, both over Europe [55.5 %] and over North.Sulfates contribute up to 44.6 % of increases in PM 2.5 over North America, and 30.0 % over Europe.The use of ULSJ fuel (FSC = 15 ppm) reduces global annual mean surface aviation-induced PM 2.5 concentrations (in relation to the NORM case) by 35.7 % [1.4 ng m −3 ] (Fig.4); predominantly due to changes in sulfate [−1.4 ng m −3 ; −62.1 %] and ammonium [−0.2 ng m −3 ; −37.9 %], which are marginally offset by very small increases in nitrates [+3.2 × 10 −3 ng m −3 ; +0.3 %].Aviation emissions also lead to small changes to other aerosol components of +0.2 ng; which includes natural aerosols such as dust [+0.3 ng m −3 ; +61.8 %], sodium [−19.5 %] and chloride from sea salt [−19.5 %] with the changes due to changes in aerosol lifetimes, along with changes in BC [−7.9 %] and OC[−19.3%].

Figure 6 .
Figure 6.Estimated global aviation-induced mortality as a function of FSC, and changes in vertical aviation emission distributions for year 2000 (Shaded region denotes the 95 % confidence through application of low-and high-range cause-specific coefficients).

Figure 8 .
Figure 8. Global and regional variations in low-cloud level (879 hPa) CCN (D p > 50 nm): (a) changes in mean concentrations and (b) percentage changes.See Fig. 5 for definitions of regions.

Figure 9 .
Figure 9. Aviation-induced radiative effects due to variations in fuel sulfur content (FSC), the ground release of aviation emissions (GROUND), and variations in the vertical distribution of aviation SO 2 emissions (SWITCH1 and SWITCH2 simulations).
estimated an O3DRE of 7.4 mW m −2 Tg(N) −1 with a model OPE of ∼ 1, while the ensemble of models considered by Myhre et al. (2011) have an OPE range of 1.5-2.4,resulting in an O3DRE range of 16.2-25.4mW m −2 Tg(N) −1 .

3. 5
Figure 10 shows the net RE and premature mortality for different aviation emission scenarios.Increases in FSC lead to approximately linear increases in both estimated mortality and the negative net RE.We quantify the impact of FSC on mortality and REs in terms of d(mortalities)/d(FSC) [mortalities ppm −1 ] and d(RE)/d(FSC) [mW m −2 ppm −1 ].We calculate the sensitivity of global premature mortality to be 1.0 mortalities ppm −1 [95 % CI = 0.4 to 1.6 mortalities ppm −1 , where the range is due to uncertainty in β].The global mean RE comb has a sensitivity of −1.2 × 10 −2 mW m −2 ppm −1 , dominated by large changes to the aCAE [−1.1 × 10 −2 mW m −2 ppm −1 ], and much smaller changes in the aDRE [−6.9 × 10 −4 mW m −2 ppm −1 ] and O 3 RE [+4.4 × 10 −5 mW m −2 ppm −1 ].The different slopes in the relationship between estimated RE and mortality (Fig.10) are driven by the range of coefficients used in the CRF.This highlights the considerable uncertainty in the health impacts caused by exposure to PM 2.5 .We note that uncertainty in the RE due to aerosol and ozone exists, but is not included in Fig.9.To assess how the vertical distributions of aviation SO 2 emissions influence human health and climate effects, we performed three additional simulations where we altered the vertical distribution of aviation SO 2 emissions (GROUND, SWITCH1 and SWITCH2 simulations).In these simulations 2 %], a larger cooling aCAE [−42.4 mW m −2 ], larger warming aDRE [2.07 mW m −2 ], resulting in additional −18.1 mW m −2 [136 %] of aviation-induced cooling [SWITCH1 RE comb of −31.4 mW m −2 ].
mW m −2 ].Z.Z.Kapadia et al.:  Impacts of aviation fuel sulfur content on climate and human health Our O3DRE [+8.9 mW m −2 ] when normalised to represent the impact of the emissions of 1 Tg(N) [+10.45 mW m −2 Tg(N) −1 ] is at the lower end of the range provided by previous studies [7.39-36.95mW m −2 Tg(N) −1 ] (Sausen

Table 2 .
FSC and global SO 2 emissions applied in each model experiment.