The Goddard Earth Observing System Model version 5 (GEOS-5) is a GCM developed by the NASA Global Modeling and Analysis Office (GMAO) . Advection is driven by the finite-volume dynamical core of , which uses sub-stepping to ensure that the Courant number does not exceed unity. Boundary layer mixing is based on the nonlocal scheme of and the Richardson-number-based scheme of . The convective parameterization is the Relaxed Arakawa–Schubert (RAS) scheme with a scheme for the generation and reevaporation of precipitation . RAS computes the effect of multiple individual cloud plumes released sequentially using a resolution-dependent stochastic trigger function . GEOS-5 has 72 vertical levels up to 0.01 hPa on a hybrid eta (sigma-pressure) grid. The horizontal grid is cubed-sphere and can operate at a range of resolutions. The integration of the model equations on the cubed-sphere grid eliminates the problem of large Courant numbers near the poles and permits straightforward domain decomposition for distributed-memory environments. The cubed-sphere grid has been used in other GCMs, such as the Geophysical Fluid Dynamics Laboratory (GFDL) Atmospheric Model 3 (AM3) .

We use here the operational GEOS-5 product (https://gmao.gsfc.nasa.gov/GMAO_products/NRT_products.php) generated at a cubed-sphere c360 (≈ 25 km) horizontal resolution. The data combines the GEOS-5 GCM with observations using a hybrid ensemble Kalman filter three-dimensional variational (3D-Var hybrid) system . The internal GCM time step for advection and convection is 7.5 min. Output from the c360 simulation is mapped onto a 0.25∘ × 0.3125∘ latitude × longitude rectilinear grid to produce the GEOS forward-processing (GEOS-FP) archive released operationally by GMAO. The archived data relevant to CTM transport and scavenging include 3-D winds, convective mass fluxes, precipitation fields, and 2-D surface pressures and boundary layer mixing depths. The 3-D data are archived as 3 h averages and the 2-D data as 1 h averages.

GEOS operational meteorological products have been widely used for off-line CTM applications including by the University of Maryland (UMD) CTM , GEOS-Chem , Global Modeling Initiative (GMI) CTM , Integrated Massively Parallel Atmospheric Chemical Transport (IMPACT) model , MOZART , CAM-Chem , and the GEOS CTM . Chemical transport simulations can also be performed online within the GEOS-5 GCM . Although we use GEOS-Chem in our comparisons against online GEOS-5 GCM results, the issues discussed in this paper are more generally pertinent to CTMs driven by archived GCM meteorological data.

We use two versions of GEOS-Chem: the standard version 11-01 released in February 2017 (http://wiki.seas.harvard.edu/geos-chem/index.php/GEOS-Chem_v11-01) and a beta high-performance version (GCHP) designed for massively parallel computing environments (http://wiki.seas.harvard.edu/geos-chem/index.php/GEOS-Chem_HP). The standard GEOS-Chem operates on the rectilinear grid from the GEOS-5 archive, while GCHP operates on the GEOS-5 cubed-sphere grid. Both versions use the same archived meteorological data and modules except for advection. GCHP uses an off-line version of the same dynamical core as GEOS-5, while the standard GEOS-Chem uses a dynamical core developed for off-line applications on a rectilinear grid . The scheme averages winds, surface pressures, and mixing ratios over the two highest latitude bands to compensate for the polar singularity. Vertical advection is computed in both cases from the change in surface pressure, but vertical advection in the standard rectilinear GEOS-Chem is lower order than in GEOS-5 and GCHP. In the standard rectilinear GEOS-Chem, a pressure fixer is used to correct for inconsistencies between horizontal wind divergence and pressure tendencies resulting from the temporal averaging in the meteorological archive, whereas GCHP enforces mass consistency by applying a global scaling factor.

Convective transport in GEOS-Chem is simulated with a single-plume scheme using the archived 3 h net updraft and detrainment convective mass fluxes summing over all RAS plumes within a given grid column . Although GEOS-Chem reproduces the bulk 3 h convective transport in the GEOS-5 GCM, the precise timing and interactions between RAS convective plumes are not resolved since that information is not in the archive. Bulk convective transport using archived mass fluxes is a standard procedure in other CTMs driven by GEOS-5 meteorology, such as the GMI CTM, and CTMs driven by other meteorology, such as TOMCAT driven by ECMWF reanalysis data.

Boundary layer mixing in GEOS-Chem uses the nonlocal parameterization of adapted for GEOS-Chem by . It draws on the archived GEOS-5 mixing depths, temperature, latent and sensible heat fluxes, and specific humidity. The mixing depths in the GEOS-5 archive are a diagnostic quantity, and thus boundary layer mixing in GEOS-Chem may differ from that in the GEOS-5 GCM.

GEOS-Chem applications typically use the native horizontal resolution of the GEOS products for nested simulations over continental-scale domains but a coarser 2∘ × 2.5∘ horizontal resolution for global simulations. The 2∘ × 2.5∘ meteorological archive is generated by averaging the original GEOS-5 archive over the corresponding grid. As part of this work, we developed a capability to conduct global GEOS-Chem simulations for passive tracers at native 0.25∘ × 0.3125∘ horizontal resolution. This allows us to separate the contributions of off-line archiving and degraded resolution to model errors.

The natural tracer suite 222Rn–210Pb–7Be provides a standard test of vertical transport and scavenging in global models, with capability to compare to observations . 222Rn is emitted ubiquitously by soils. Its sole sink is radioactive decay to 210Pb with a half-life of 3.8 days, making it a sensitive tracer for vertical transport in the troposphere . 210Pb (half-life 22.3 years) attaches to aerosol particles and provides a diagnostic for aerosol lifetime against deposition . 7Be (half-life 53.3 days) is produced in the upper troposphere and lower stratosphere from the interaction of cosmic rays with atmospheric oxygen and nitrogen and attaches to aerosol particles and is removed by deposition in the same way as 210Pb. The high-altitude source of 7Be complements 210Pb by testing the model representation of subsidence and stratosphere–troposphere exchange .

Here we use the 222Rn–210Pb–7Be simulation originally developed for GEOS-Chem by . 222Rn is emitted uniformly from land excluding ice at a rate of 1.0 atoms cm-2 s-1 under nonfreezing conditions and 0.3 atoms cm-2 s-1 under freezing conditions. The 7Be source function depends only on altitude and latitude. 210Pb and 7Be are removed by aerosol wet and dry deposition, in addition to radioactive decay (negligible for 210Pb). Dry deposition is a minor sink. Wet deposition includes scavenging in convective updrafts following the approach of , and first-order in-cloud and below-cloud scavenging for anvil and large-scale (grid-resolved) precipitation following the approach of . Aerosol can be released below cloud if precipitation evaporates. The scavenging parameterizations of are intended to be applicable to GEOS-Chem at all resolutions because convective mass fluxes (from the GEOS-5 archive) do not change with resolution and because first-order rainout or washout assumes precipitating fractions of grid boxes that are set in all cases by a fixed rate of conversion of cloud water to precipitation .

A recent study by uses GEOS-Chem with an updated 222Rn source function to evaluate with observations worldwide. Here our focus is not on comparison to observations but on the effect of CTM model differences relative to a reference simulation. Since the averaging and remapping of meteorological fields in the CTM represents a degradation of the information from the reference simulation, we view for our purpose the reference simulation as the “truth” against which the different CTM simulations can be compared.

We conducted a number of online and off-line 222Rn–210Pb–7Be simulations for different spatial resolutions and configurations, as illustrated in Figure  and explained below. All simulations were conducted for 2 months starting from zero concentrations on 1 June 2013. We report and compare monthly mean results for July 2013 after a 1-month (June) spin-up. We limited the analysis to 1 month because of computational and storage requirements for the high-resolution simulations and with the expectation that 1 month in Northern Hemisphere summer is a sufficient time window to diagnose systematic differences in vertical tropospheric transport as revealed by the 222Rn–210Pb–7Be system. The zero initialization allows for sensitive analysis of differences but implies that concentrations are not in steady state and should not be compared to observations, in particular for 210Pb and 7Be. For example, observed and steady-state GEOS-Chem 210Pb concentrations in the stratosphere are higher than in the troposphere , but in our simulations they are much lower. Stratospheric concentrations of 222Rn and 210Pb should not be compared across simulations since the injection of 222Rn into the stratosphere may be driven by sporadic deep convection . The stratosphere is not discussed in what follows.

Simulation 1 is conducted online using a version of GEOS-5 similar to the one used in at c360 resolution. For 222Rn, it defines the reference simulation; all other simulations in Fig.  (blue boxes) successively degrade some aspect of that reference simulation. This is not the case for 210Pb and 7Be because the scheme for aerosol wet scavenging in the GEOS-5 GCM is less advanced than in GEOS-Chem. In particular, aerosol scavenging in the GCM is not coupled to sub-grid transport in deep convective updrafts, and this can severely overestimate the transport of aerosols to the upper troposphere . Thus we do not show 210Pb and 7Be results from the online simulations.

Simulation 2 is also conducted online at c360 resolution but with the bulk convective algorithm of GEOS-Chem and 3 h averaged convective mass fluxes from simulation 1. This allows us to separately examine the effect of using archived fields on convection and advection. Simulation 2 is used to generate the meteorological archive (3 h for winds and convective mass fluxes, 1 h for mixing depths) for the off-line GEOS-Chem simulations. This off-line archive mimics the operational GEOS-FP archive by using the same temporal averaging windows and remapping the cubed-sphere meteorological data to a 0.25∘ × 0.3125∘ rectilinear grid.

Simulation 3 is the standard off-line high-resolution GEOS-Chem on a 0.25∘ × 0.3125∘ rectilinear grid. It uses an archive of winds, mixing depths, and convective mass fluxes generated from simulation 2 that mimics the GEOS-5 operational product. Errors in the 222Rn simulation compared to simulation 2 include the temporal averaging of winds and mixing depths, the remapping of the meteorological archive to a rectilinear grid, and the use of a lower-order advection core and a different boundary layer mixing scheme. For 210Pb and 7Be, simulation 3 represents our best-case reference simulation.

Simulation 4 is the standard off-line coarse-resolution GEOS-Chem on a 2∘ × 2.5∘ grid. It uses the off-line meteorological archive from simulation 2 but degraded to 2∘ × 2.5∘ resolution. Comparison with simulation 3 shows the error from degraded horizontal resolution. Comparison to simulation 1 (for 222Rn) shows the compounded errors in going from the original online GEOS-5 simulation to the off-line, coarse-resolution simulation.

Although global GEOS-Chem simulations may be conducted at coarse 2∘ × 2.5∘ resolution, they use driving meteorological fields generated from the original GEOS-5 simulation at c360 (≈ 0.25∘ × 0.3125∘) resolution. This is an important distinction from a simulation that would be driven by a coarser meteorological model. To investigate that effect, we also conducted an online simulation 5 using GEOS-5 meteorology at c48 resolution (≈ 2∘ × 2.5∘). We then used the meteorological data from simulation 5, archived on the cubed-sphere grid, to drive an off-line c48 simulation using the high-performance version of GEOS-Chem on that grid (simulation 6) and an off-line 2∘ × 2.5∘ simulation using the standard GEOS-Chem on a rectilinear grid (simulation 7). A comparison of simulations 6 and 7 diagnoses the error from remapping the meteorological archive from its native cubed-sphere to a rectilinear grid. Except for temporal averaging, the off-line archive for simulation 6 is fully consistent with the c48 online simulation 5 (no remapping). The c48 resolution allowed us to conduct a cubed-sphere off-line (GCHP) simulation, which we were not able to do at c360 resolution due to computational limitations.

Together, simulations 1–7 allow us to examine and isolate different sources of error in simulations of chemical transport including meteorological grid resolution, off-line meteorological archiving (temporal averaging), remapping of the meteorological archive, spatial degradation of that archive, and differences between off-line and online transport schemes. Salient results are discussed in the next section. We use monthly average zonal mean profiles vs. altitude and latitude as our comparison metric, following standard practice for 222Rn model intercomparisons . Another comparison metric for 210Pb and 7Be is the global tropospheric lifetime against deposition . Throughout this paper, we refer to “archiving” as the temporal averaging of meteorological fields for use in off-line simulations, “remapping” as the cubed-sphere to rectilinear transformation of these fields, and “spatial averaging” as the further degradation of these fields from a fine to a coarse off-line grid.

Figure b shows the percentage differences in zonal mean 222Rn profiles between simulation 2 (off-line GEOS-Chem convection scheme) and simulation 1. Simulation 2 has up to 10 % higher 222Rn concentrations in the equatorial lower troposphere and up to 7 % lower 222Rn concentrations in the middle to upper troposphere. The combination of using the GEOS-Chem convection scheme and temporally averaged convective mass fluxes results in slightly reduced vertical transport compared to the original GEOS-5 convection.

The GEOS-Chem convection scheme operates as a single convective plume in each grid column on the basis of the 3 h archive of GEOS-5 convective updraft and detrainment data. We find in a sensitivity simulation that using 15 min or 3 h averages of convective mass fluxes makes no significant difference. Thus the differences in Fig.  arise mainly from the bulk convective transport scheme used in GEOS-Chem, which simplifies the RAS ensemble-plume parameterization to a single plume. One explanation for why a multi-plume parameterization might produce a different transport pattern is that each sequential plume acts on a different concentration gradient that has been modified by the previous plume until the moisture and temperature fields are balanced. A tall plume followed by a series of short plumes will transport more tracer higher than a series of shorter plumes followed by a tall plume.

Figure c shows the percentage differences between simulation 3 (off-line) and simulation 2 (online). The off-line simulation has higher 222Rn concentrations in the mid-troposphere (700–500 hPa) and lower concentrations in the upper troposphere (above 500 hPa), with some differences exceeding 20 %. Encapsulated in this comparison are the effects of remapping the archived meteorological fields to a rectilinear grid using a different advection scheme, using a different boundary layer mixing scheme, and using 3 h averaged wind fields. Large relative differences in polar grid cells may reflect the averaging of concentrations in the polar latitudes in the rectilinear advection scheme and the transition to semi-Lagrangian advection when the Courant number exceeds unity. 222Rn concentrations are very low in these polar grid cells so that absolute differences are small.

To better understand the contributions from different sources of error in the off-line simulation, we examine simulations 5–7, which show a similar transition from online to off-line, but starting at c48 resolution in the GEOS-5 model and with the intermediate addition of a c48 off-line GCHP simulation (custom cubed-sphere archive, no remapping). In this way we can diagnose the effects of using archived meteorology separately from the effects of the advection core and remapping error associated with conversion to a rectilinear (here 2∘ × 2.5∘) grid. Figure a shows the zonal mean 222Rn concentrations for c48 resolution, which have a similar pattern to the c360 results. Figure b and c additively separate out the contributions from off-line archiving (simulation 6 vs. simulation 5) and the remapping and transport scheme (simulation 7 vs. simulation 6). Off-line archiving results in overall weaker vertical transport, as might be expected from transient motions averaged out in the meteorological archive. The bias is about 5 % in surface air but can exceed 20 % in the upper troposphere. There is still a bias over Antarctica even though the off-line cubed-sphere geometry does not have a polar singularity; this may reflect the cumulative effect of meridional transport differences affecting a region particularly remote from sources. The combination of remapping to the rectilinear grid and using the advection scheme (right panel) also incurs differences of about 5 % in surface air and up to 20 % in the upper troposphere. We would expect to see larger remapping errors associated with smaller grids, especially over the polar regions. The errors from simulations 6 and 7 compound for surface air but tend to cancel in the upper troposphere.

Figure  shows the effect of grid resolution on zonal mean 222Rn profiles in GEOS-5 (online, with resolution affecting meteorology) and GEOS-Chem (off-line, using the same meteorological archive in both cases). Figure 4b shows the percentage difference between simulation 5 (online c48) and simulation 2 (online c360). The c360 meteorological simulation has sufficient spatial resolution to resolve large convective systems and thus has much less parameterized convection than the c48 simulation. The higher 222Rn concentrations in the tropical upper troposphere in c48 and lower concentrations in the extratropical upper troposphere most likely reflect differences in vertical transport properties between the resolved and parameterized convective formulations. Thus the higher concentrations in the tropical middle to upper troposphere at c48 can be attributed to the convective fluxes that are not represented by the resolved transport at c360. The lower concentrations near the tropopause can be attributed to the known insufficiency of convective transport for reaching that level . The lower concentrations in the extratropical upper troposphere may be due to inability to diagnose organized vertical motion as convection. The higher concentrations over Antarctica at c48 may be due to numerical diffusion during the slow long-range meridional transport from lower latitudes . While these differences are large, absolute concentrations are very low over the poles due to the short lifetime of 222Rn and lack of emission over ice and snow.

Degrading model resolution has different effects in GEOS-Chem, in which the coarse 2∘ × 2.5∘ simulation uses the same 0.25∘ × 0.3125∘ meteorological archive as the high-resolution simulation but with spatial averaging of the meteorological fields. The coarse simulation results in decreased vertical transport to the upper troposphere at all latitudes, with maximum effect (up to 40 %) in the subsiding subtropics. This may be simply explained by the averaging out of vertical eddy motions on the coarser grid, including organized vertical motions across multiple levels that the online c48 simulation would simulate with stronger parameterized convection. This systematic bias in the coarse-resolution GEOS-Chem may thus be correctable through the addition of convective motions. We explore this idea in the next section.

Figure  (top panels) shows the zonal mean concentrations of 210Pb and 7Be from the GEOS-Chem simulation at 0.25∘ × 0.3125∘ (simulation 3). The 210Pb distribution is shifted to higher altitudes relative to 222Rn, reflecting the effect of scavenging in the lower troposphere. 7Be shows preferential subsidence in the dry subtropics and is depleted in the lower troposphere by scavenging. 7Be concentrations are low throughout the tropical troposphere due to dominant upwelling of 7Be-depleted surface air.

Mean tropospheric lifetimes against deposition in the 0.25∘ × 0.3125∘ simulation are 6.7 days for 210Pb and 17 days for 7Be. previously inferred 210Pb and 7Be tropospheric residence times of 9 and 21 days, respectively, from an earlier version of GEOS-Chem evaluated with observations. A more recent evaluation by using GEOS-Chem version 11-01 and an updated 222Rn source finds that a residence time for 210Pb of 7 ± 1 days better matches the observational constraints.

The bottom panels show the effects of degrading the GEOS-Chem resolution to 2∘ × 2.5∘. Overall the results are consistent with our previous finding for 222Rn that degrading the resolution weakens vertical transport. Tropospheric lifetimes decrease to 6.2 days for 210Pb and increase to 18 days for 7Be, with is consistent with the shifts in vertical distribution to the lower troposphere for 210Pb and to the upper troposphere for 7Be. 210Pb concentrations are higher in the tropical upper troposphere in the 2∘ × 2.5∘ simulation, likely due to differences in wet scavenging.