Historical anthropogenic emissions in CMIP6 are from the Community Emissions Data System CEDS,. For surface emissions, we processed version 2017-05-18 of the CEDS inventory available at monthly temporal and 0.5∘ spatial resolution for the eight surface sectors listed in Table  and the 1850–2014 CE period. CEDS is presently the default global surface anthropogenic emissions inventory used by GEOS-Chem, although only species for the full-chemistry simulation have been processed, and these emissions are overwritten for many locations by regional inventories. Here, we additionally processed the methane and CO2 fluxes for those GEOS-Chem specialty simulations. All surface emissions increased exponentially over the historical period except for sulfur dioxide (SO2), whose emissions peaked in the 1980s (Fig. S52). We also processed the three-dimensional CEDS aircraft emissions for input into GEOS-Chem as a CMIP6-compliant alternative to the Aviation Emissions Inventory Code AEIC, source that is the default in GEOS-Chem. We processed version 2017-08-30 of the CEDS aircraft inventory, available for NO, CO, black and organic carbon, SO2, and ammonia (NH3) at monthly temporal and 0.5∘ spatial resolution and 25 vertical levels of equal thickness from the surface to 15 km. We have vertically regridded to the native 40-level E2.1 resolution and the reduced stratospheric 47-level MERRA-2 resolution. Global aircraft emissions increased mostly linearly across the historical period beginning in approximately 1950 CE (Fig. S53).

In addition, we processed the biomass burning emissions from version 1.2 of the Biomass Burning for Model Intercomparison Projects (BB4MIPs) inventory . The BB4MIPs reconstruction combines version 4 of the satellite-based Global Fire Emissions Database GFED4;the default biomass burning inventory in GEOS-Chem and available since 1997 CE with observational proxies and model simulations from the Fire Model Intercomparison Project (FireMIP) for earlier periods. BB4MIPs provides the total mass flux per species at monthly temporal and 0.25∘ spatial resolution. We have regridded to 0.5∘ spatial resolution and speciated for input to GEOS-Chem using a consistent hydrocarbon speciation scheme with CEDS. Fire emissions from BB4MIPs increase slightly across the historical period. However, interannual variability greatly increases in the second half of the 20th century (Fig. S54), driving the large interannual variability observed in the total emissions during this period (e.g., Fig. b).

We have prepared version 1.2.0 of the CMIP6 surface boundary conditions derived from historical observations and available at monthly temporal resolution and as 0.5∘ latitude bands. These have been regridded to 0.5∘ global spatial resolution for input into GEOS-Chem. Carbon dioxide, methane, and nitrous oxide have monotonically increased since the pre-industrial era due to anthropogenic activity, whereas shorter-lived stratospheric ozone-depleting substances peaked around the turn of the last century following their ban under the Montreal Protocol (Fig. ).

The future anthropogenic emissions and boundary conditions used by CMIP6 and processed here for GCAP 2.0 are summarized by . In brief, the so-named Shared Socioeconomic Pathways (SSPs) are determined from integrated assessment modeling (IAM) of five future societal narratives that may be followed to limit future warming to a target radiative forcing. The nomenclature for characterizing the SSP scenarios is SSPx-y.z (or SSPxyz), where x is the number of the future narrative (1 to 5; Table ), and y.z represents the target radiative forcing in W m-2 at the end of the 21st century ranging from 1.9 (extreme mitigation; low warming) to 8.5 (low mitigation; extreme warming). The names of the five narratives are listed in Table , and each employs different assumptions about how global society would achieve the target radiative forcing. SSP1 assumes low challenges to mitigation and adaptation , SSP2 assumes medium challenges to mitigation and adaptation , SSP3 assumes high challenges to mitigation and adaptation , SSP4 assumes low challenges to mitigation and high challenges to adaptation , and SSP5 assumes high challenges to mitigation and low challenges to adaptation .

For GCAP 2.0, we focus on seven scenarios corresponding to Tiers 1 and 2 of the ScenarioMIP experiment: SSP1-1.9, SSP1-2.6, SSP4-3.4, SSP2-4.5, SSP4-6.0, SSP3-7.0, and SSP5-8.5 . Assumptions about future population growth, urbanization, gross domestic production (GDP), energy, land use, and air pollution trends of the SSP IAM scenarios are described in a series of papers . Version 1.1 of these scenarios was obtained from input4MIPs on ESGF and available at 0.5∘ spatial and monthly resolution for 2015 CE and then every 10 years beginning with 2020 CE and ending with 2100 CE. These future emissions were processed for input to GEOS-Chem in the same way as the historical emissions and boundary conditions. Linear interpolation was used to develop individual yearly emissions between the available decadal values. In addition to the sectors of Table , future CO2 emissions include a “neg” sector that considers negative emission (i.e., carbon capture).

From the perspective of the simulated climate, the forcing that dominates the end-of-the-century response is the CO2 abundance. Therefore, CO2 is the only gas that monotonically increases with future forcing target values (Fig. a). The trajectories of the remainder of the well-mixed greenhouse gases, stratospheric ozone-depleting substances, short-lived climate forcers, and air-pollution precursors vary between the SSP scenarios and target forcings (Figs.  and b–l). For example, methane in 2100 CE is highest in the SSP3-7.0 scenario, not SSP5-8.5 (Fig. b). Because ammonia is primarily produced from agriculture and the world population will continue to grow, it is the only emission expected to remain constant or grow into the future. Otherwise, the narrative assumptions strongly influence the global and regional emission changes. Furthermore, we note that the historical emissions inventories have large amounts of interannual variability (primarily due to biomass burning), whereas the SSP scenarios have very low interannual variability. This highlights the necessity for CTM studies such as those that may be accomplished by GCAP 2.0 that can explore the impact of a wider range of future emission trajectories on air quality, short-lived climate forces, and stratospheric ozone in future warmer climates whose meteorology is primarily driven by changes in CO2. It also highlights the necessity for performing simulations long enough to establish robust statistics for chemistry–climate interactions e.g.,.

Figure  shows observed meridional and vertical gradients of SF6 with respect to Cape Matatula, American Samoa (SMO), in the remote tropical southern Pacific. The observations are version 2.1.1 of the NOAA Carbon Cycle Group SF6 ObsPack (10.25925/20180817, ) and represent a mixture of surface in situ, flask, tower, and aircraft sources from 2005–2014 CE. Observations were aggregated at model spatial and monthly temporal resolution, compared to that month's SMO value, from which zonal climatologies were determined. Also shown is the value of each simulation sampled and processed as in the observations. In all simulations, GEOS-Chem underestimates the cross-equatorial meridional gradient by 17 %–26 %, an improvement over GEOS-Chem driven by earlier meteorology products see Supplement of. However, this suggests that the interhemispheric mixing rate in the model is too fast and/or that the EDGAR inventory underestimates the emission growth rate in the Northern Hemisphere relative to the Southern Hemisphere. Meanwhile, meridional mixing rates in the Southern Hemisphere are consistent with the observations in all simulations. The E2.1 simulation slightly better matches the cross-equatorial gradient than the MERRA-2 and nudged simulations, but E2.1 greatly underestimates the lower stratospheric gradient (see Sect. ).

We assess vertical mixing within the troposphere using vertical profiles of 222Rn and the ratio of 7Be to 210Pb in surface air.

Figure  shows simulated climatological 222Rn profiles sampled at the month and location of the available observations, also plotted. Observations are scarce and available only at northern extratropical continental locations . In an overly convective atmosphere, the vertical gradient of 222Rn would disappear. There is a slight overestimate of 222Rn abundance within the boundary layer and underestimate above in all of our simulations, implying a small underestimate in boundary layer ventilation. Our results are comparable to or better than other atmospheric models e.g., see Fig. 5 of.

Figure  shows the annual mean surface mixing ratios of 7Be, 210Pb and their ratio in our simulations. Since 7Be is produced in the upper troposphere/lower stratosphere and 210Pb is produced near the surface, and because the ratio of 7Be to 210Pb is unaffected by deposition, the ratio serves as a useful measure for tropospheric vertical mixing . A persistent high bias would indicate excessive downward transport and/or insufficient upward transport, assuming no bias in either source. The left column shows the climatological long-term data from the surface monitoring stations of the DOE Environmental Measurements Laboratory (EML) Surface Air Sampling Program (SASP) (obtained from https://www.wipp.energy.gov/namp/emllegacy/databases.htm, last access: 15 January 2021​​​​​​​). SASP recorded the spatial and temporal distribution of various radionuclides in surface ambient air from 1957 until 1999 CE, including 7Be and 210Pb. For comparison with our simulations, we select data from periods of average solar activity (solar modulation potential Φ = 670 ± 50 MV from the reconstruction). All simulations show only minor biases in surface abundance in either 7Be and 210Pb. The 7Be to 210Pb ratio shows tight agreement between the different model simulations, with the E2.1 simulation slightly outperforming the MERRA-2-driven simulation relative to the observations.

Figure  shows the simulated zonal climatology of the age of air in the stratosphere, which is defined as the mean time since an air mass at a given location was last in the troposphere . Age of air increases away from the equatorial tropopause where most tropospheric air enters the stratosphere . We determine age of air by using SF6 as a chronological tracer e.g., to determine the average temporal lag between a mixing ratio at a given location in the stratosphere relative to the tropical tropopause for the period 2005–2014 CE.

Models traditionally underestimate the stratospheric age of air implied by observations, which increases up to 7 years by 35 km in the poles . All three of our simulations, including the MERRA-2 reanalysis, underestimate the age of air. E2.1 has the youngest air, with values consistent with those determined online within the GCM and reported by . E2.1 nudged to MERRA-2 has the oldest air but is still only about two-thirds of what observational constraints suggest they should be and remain difficult to interpret given the challenges of nudged simulations e.g.,. The young age in E2.1 results from too strong an ascent in the tropical pipe and a relatively leaky lower branch of the Brewer–Dobson circulation e.g.,; updates to version E2.2 of the GISS GCM greatly improve the stratospheric circulation and will be included in future GCAP 2.0 meteorology products and scenarios.

Beryllium-7 has often been used as a tracer of downward transport from the stratosphere and as an indicator of stratosphere–troposphere exchange (STE) performance within global atmospheric models .

Figure  shows the annual zonal fraction of 7Be of stratospheric origin in each simulation. Using E2.1 (MERRA-2) meteorology, we find that 23 % (30 %) of annual average surface 7Be abundance from 38–51∘ N is of stratospheric origin, slightly lower (higher) than the observational constraint of 25 % reported by . This is a dramatic improvement over the earlier GCAP/ICECAP studies in which the stratospheric downwelling source in the E2.1-driven simulations was greatly overestimated and was also seen in the NASA Global Modeling Initiative (GMI) CTM driven by GISS Model II' meteorology . This possibly reflects improvements in downward mass flux in the GCM associated with the increase in vertical resolution (Fig. ) and updates to the GEOS-Chem dynamical core code that occurred since the original GCAP/ICECAP.

Figure  (top row) compares the simulated wet deposition flux of 210Pb in each simulation to the observational values reported by aggregated to model resolution. Wet deposition of 210Pb in all simulations is biased low by ∼ 20 % in all simulations, consistent with earlier versions of GEOS-Chem . These results imply a low bias in the 210Pb source (and therefore 222Rn emission) or a high bias in the dry deposition flux in all simulations.

Figure  (bottom row) compares the simulated wet deposition fluxes of 7Be to the few observations that exist and references therein. The wet deposition flux is increased in the E2.1 simulation with respect to the MERRA-2 simulation, particularly over the midlatitude oceans, and is the closest simulation to matching the observations.

We find a tropospheric residence time for 210Pb-containing aerosols against deposition of 8.3 d in MERRA-2 versus 8.8 d in the E2.1 simulations (Table ), an improvement in consistency with respect to GCAP/ICECAP, and within the range of previous estimates of 6.5–12.5 d . We find a similar relative increase in the lifetime of 7Be-containing aerosols of 19.7 d versus 21.7 d, also consistent with earlier findings of 23 d and 21 d . Whereas wet deposition in MERRA-2-driven simulations is primarily due to stratiform clouds, it is primarily due to convective clouds in E2.1-driven simulations, consistent with the meteorological fields (see Sect. ) and leading to the longer residence times in the E2.1 simulations.

Table  assesses hydroxyl radical (OH) concentrations in the model by comparing the simulated lifetime of relatively long-lived molecules whose main sink is tropospheric OH with observational constraints. It is common for atmospheric models to be biased high with respect to OH in these observational constraints i.e., low with respect to lifetime, e.g.,, which is true as well in our three simulations. However, the lifetime of methyl chloroform (CH3CCl3) and methane in the E2.1 simulations is statistically consistent with the low end of the observed constraint of 6.0-0.4+0.5 years and 10.2-0.7+0.9 years from , respectively. All simulations fall within the multi-model estimates of a tropospheric methane lifetime of 10.2 ± 1.7 years and 9.8 ± 1.6 years but fall short of the observationally derived estimates of 11.2 ± 1.3 years from . In all simulations, the E2.1 simulations better match the observational constraints. The E2.1-driven simulations in the tropics have thinner overhead ozone columns (see Sect. ), greater free-tropospheric water vapor abundances, and greater lightning NO emissions (see Fig. ), all of which promote increased OH e.g.,, so their lower OH may reflect their higher biogenic NMHC emissions (Fig. ).

Table  gives the tropospheric budget for total reactive nitrogen (NOy) in all three simulations, which includes NOx (≡ NO + NO2) and its longer-lived reservoir species such as nitric acid (HONO2) and peroxyacetyl nitrate (CH3C(O)OONO2; PAN). The E2.1 simulations have greater total sources of NOy because of the larger natural sources from lightning and soils (see Sect. ) and a greater flux of NOy transported from the stratosphere from the products of nitrous oxide (N2O) oxidation. The NOy speciation between family members is largely consistent between the simulations. The lifetime of NOy in the nudged E2.1 simulation is longer than in the free-running E2.1 simulation, reflecting the reduction in that simulation's global mean precipitation flux.

Panels a–d of Fig.  compare the spatial distribution of tropospheric columns of nitrogen dioxide (NO2) for 2005–2014 CE from version 3 of the OMNO2d product from the Ozone Monitoring Instrument (OMI) on the Aura satellite (10.5067/Aura/OMI/DATA3007, ) to the three simulations. Figure S61 in the Supplement provides seasonal details. The simulations have been sampled at the satellite's overpass time and the tropopause was determined online within the model following the thermal definition is used to calculate the partial columns. All simulations well reproduce the spatial distribution of tropospheric NO2 (all R≥0.94) and have small global mean low biases but statistically disagree with the satellite product in the subtropical latitudes. The subtropical disagreement is potentially due to uncertainties in the tropopause height between the various products. Tropospheric columns match over East Asia during this period but are underestimated over North America and Europe.

Table  gives the tropospheric emissions and lifetimes for key reduced carbon species in all three simulations. Species primarily emitted from terrestrial plants such as isoprene and monoterpenes have higher emission rates in E2.1 due to the higher diffuse radiation fluxes than in MERRA-2 (see Sects.  and ). Species that are primarily lost via oxidation by OH, such as isoprene, have shorter lifetimes in E2.1 due to the higher OH abundances (see Table ). In contrast, soluble species such as methanol have longer atmospheric lifetimes in E2.1 due to the lower large-scale stratiform precipitation rates in E2.1 relative to MERRA-2.

Panels e–h of Fig.  compare total columns of formaldehyde (HCHO) from version 3 of the OMHCHOd product from OMI on the Aura satellite (10.5067/Aura/OMI/DATA3010, ) to the three simulations for 2005–2014 CE. Figure S62 in the Supplement provides seasonal details. The simulations have been sampled at the overpass time of the satellite. Formaldehyde is a common product of hydrocarbon oxidation. Despite higher mean values over the continents, all three simulations are statistically consistent with respect to the large amount of interannual variability in the simulated and observed HCHO columns. There is strong spatial correlation between the simulations and the satellite product (all R≥0.8). Terrestrial columns simulated by the E2.1 simulations are higher than the MERRA-2 simulations, reflecting the higher biogenic emissions. All simulations underestimate HCHO columns over the remote ocean aside from the continental outflows of North America and Asia and over the Arctic.

Table  gives the tropospheric budget for carbon monoxide (CO) in all three simulations. The direct emissions of CO from anthropogenic and biomass burning sources between the three simulations are identical by experimental design. However, the chemical source of CO from methane and non-methane hydrocarbon oxidation is higher in the E2.1 simulations due to the higher OH abundances (Table ). This is balanced by the increased chemical loss of CO by OH. The influence of OH on CO production from short-lived non-methane hydrocarbon species seems to outweigh the influence on CO loss, and consequently there are slightly longer CO tropospheric lifetimes in the E2.1 simulations.

Panels i–l of Fig.  compare CO mixing ratios at 500 hPa from version 7 of the AIRX3STD product from the Atmospheric Infrared Sounder (AIRS) on the Aqua satellite (10.5067/8XB4RU470FJV, ) to the three simulations for 2005–2014 CE. Figure S63 in the Supplement provides seasonal details. The simulations have been sampled at the overpass time of the satellite. All simulations have strong spatial correlation with the observations (R=0.93). CO in the free troposphere is higher in the E2.1 simulations, consistent with its longer lifetime (Table ). The E2.1 simulations are globally biased high by 15 % compared to the AIRS values. However, they are statistically consistent almost everywhere with respect to the large amount of interannual variability in the observations and simulations. In contrast, the MERRA-2 simulations are globally biased low by 5 % and significantly so throughout the tropics and Southern Hemisphere. All models underestimate the curious enhancement of CO seen in the AIRS climatology over Antarctica.

Table  gives the tropospheric budget for the odd-oxygen family (Ox ≡ O3 + rapid cycling species) for all three simulations. The individual budget terms are all are consistent with the range of reported values from the Tropospheric Ozone Assessment Report (TOAR) multi-model assessment see Fig. 3 of, the CMIP6 models that performed interactive tropospheric chemistry , as well as the last extensive tropospheric ozone budget evaluation within the standard GEOS-Chem model . The E2.1 simulations are on the high end of the previously reported values due to the higher tropopause height in those simulations; the upper troposphere and lower stratosphere regions contribute strongly to each Ox budget term due to the rapidly increasing abundances of ozone with altitude there. We point out that the stratosphere-to-troposphere flux calculated using the “residual method” of the other budget terms yields consistent results when we track the mass flux of ozone across the dynamic tropopause in the model. Relative to the earlier GCAP and ICECAP simulations , the transport of ozone from the stratosphere is dramatically improved.

Figure  evaluates the zonal and seasonal distribution of ozone versus in situ measurements. We use the ozonesonde measurements archived by the World Ozone and Ultraviolet Radiation Data Centre (WOUDC) of the World Meteorological Organization/Global Atmosphere Watch Program (WMO/GAW). The data were accessed on 4 November 2019 from 10.14287/10000001 . All models fall within the variability of the measurements, except the free troposphere of the northern extratropics, where the simulations are biased low, especially during the summer months. The MERRA-2 simulations better reproduce the seasonality of ozone in the tropical free and upper troposphere, likely reflecting the climatologically constrained lightning NO source in that version (see Sect. ).

Figure  evaluates spatial distributions of ozone in the three simulations against surface in situ and satellite observations. The top row shows the annual average surface ozone mixing ratio in ppbv (≡ nmol mol-1) from our simulations versus the gridded mean non-urban surface ozone product from the Tropospheric Ozone Assessment Report (TOAR) (10.1594/PANGAEA.876108, ). The middle row shows tropospheric columns of ozone (TCO; in Dobson units) versus the joint Ozone Monitoring Instrument (OMI) and Microwave Limb Sounder (MLS) product from the Aura satellite . The bottom row shows total ozone columns (TOCs; in Dobson units) versus the OMDOAO3e product from OMI (10.5067/Aura/OMI/DATA3005, ). All observational products have been aggregated to model resolution for comparison. In the case of the satellite products, we sampled the model at the overpasses' time and location. We use the online thermal lapse rate tropopause to calculate TCO. Seasonal details are given in Figs. S64–S66 of the Supplement.

All simulations are biased high by 15 %–17 % with respect to surface ozone, mostly driven by the eastern North American data. The E2.1 simulations are especially higher over the Amazon than either MERRA-2 or the observations. Comparisons of TCO to OMI/MLS are sensitive to uncertainties in the tropopause location in the satellite product versus the models . The E2.1 TCOs are globally higher than the OMI/MLS product by 12 %, reflecting the lower tropopause pressures. However, the only locations in which it is statistically different regarding interannual variability are over the tropical oceans, where it is biased low. This likely reflects the more vigorous convection in E2.1 that leads to ozone destruction e.g.,. Tropospheric columns in MERRA-2 match the global mean from OMI/MLS but also underestimate the western Pacific and additionally underestimate northern extratropical ozone. All simulations underestimate tropospheric columns in the southern extratropics with respect to OMI/MLS. The models all show excellent agreement with respect to total ozone columns with small positive mean global biases of 4 % and high pattern correlation (all R≥0.96). Total ozone in the tropics is higher in MERRA-2 than E2.1, consistent with differences in the rate of vertical ascent in the tropical pipe implied by the stratospheric age of air comparison (see Sect. ). The E2.1 simulation overestimates Antarctic ozone relative to MERRA-2 or the nudged simulation, although not significantly compared to interannual variability.

Figure  evaluates the spatial distribution of particulate matter in the three simulations against satellite observations. The top row shows the simulated concentration of fine particulate matter under 2.5 µm (PM2.5) in µg m-3 versus the historical reconstruction from . The middle row shows the total aerosol optical thickness (AOT) at 550 nm (unitless) in the simulations versus the combined Dark Target and Deep Blue AOT at 0.55 µm from Collection 6.1 of the MODerate resolution Imaging Spectroradiometer (MODIS) MYD08 product from the Aqua satellite (10.5067/MODIS/MYD08_M3.061, ). The bottom row shows the total column of sulfur dioxide (SO2) in Dobson units in the simulations versus the second public release of version 3 of the OMSO2e product from OMI on the Aura satellite (10.5067/Aura/OMI/DATA3008, ). The model has been sampled at the time of the Aura and Aqua overpasses for comparison to the satellite products. Figures S67–S68 of the Supplement provide seasonal details for AOT and the SO2 columns.

Surface PM2.5 agrees between the simulations and over heavily industrialized regions. However, we note that the product used ratios of surface PM2.5 to AOT from GEOS-Chem to generate their proxy reconstruction from satellite AOT measurements, so it is not an entirely independent comparison. Surface concentrations are underestimated almost everywhere else in all simulations, especially in regions heavily influenced by mineral dust and biomass burning. The same story is reflected in the MODIS AOT, except the stronger sea-salt aerosol emission in the E2.1 simulations yielding better comparison in the Southern Ocean and portions of the Pacific. Simulated total columns of sulfur dioxide are underestimated everywhere in all simulations except over East Asia and the locations of the major volcanic eruptions of 2005 (Sierra Negra in the Galápagos and Anatahan of the Northern Mariana Islands).