Stratosphere–troposphere exchange (STE) of air masses and chemical species occurs at small, synoptic and global scales. It is typically associated with the occurrences of tropopause folding and cutoff cyclones and, more importantly, the global circulation of the atmosphere (Holton et al., 1995). While stratosphere-to-troposphere transport removes many chemical species from the stratosphere, it represents a significant source of ozone and other reactive species for the tropospheric chemical system (Stohl et al., 2003). Ozone is an important greenhouse gas, especially in the upper troposphere. It is a harmful pollutant near the surface where stratospheric ozone intrusions may make significant contributions (e.g., Lin et al., 2012, 2015; Langford et al., 2015). It is also the main precursor of hydroxyl radicals (OH) and thus plays an essential role in the oxidizing capacity of the troposphere. In a warmer climate, the stratosphere may increase its contribution to tropospheric ozone levels due to a stronger residual circulation (Collins et al., 2003). Quantitative understanding and prediction of anthropogenic (vs. natural) perturbations to tropospheric ozone require the use of global 3-D models; correctly representing the STE flux in these models is therefore critical. However, current models show large (30 %) uncertainty in predicted STE fluxes of ozone (Stevenson et al., 2006). Here we use the Global Modeling Initiative (GMI) modeling framework (Douglass et al., 1999; Rotman et al., 2001) to assess the utility of the aerosol tracer beryllium-7 (7Be) for evaluating cross-tropopause transport in global models.

Beryllium-7 has a half-life of 53.3 days and is produced by cosmic ray spallation reactions in the stratosphere and upper troposphere. After production, it attaches immediately to ubiquitous submicron aerosols in the ambient air. The fate of 7Be then becomes that of those aerosols, which move with the air until scavenged by precipitation or deposited to the surface. 7Be is a useful aerosol tracer for testing wet deposition processes in a global 3-D model and is often used in conjunction with the terrigenic 210Pb aerosol tracer, as wet deposition is its principal sink and its sources are relatively well known (e.g., Brost et al., 1991; Koch et al., 1996; Liu et al., 2001). On the other hand, because of its source at high altitudes and the large concentration vertical gradient, simulation of 7Be tests the model's capability to describe stratosphere-to-troposphere transport and subsidence in the troposphere (e.g., Liu et al., 2001; Allen et al., 2003).

Beryllium-7 has long been recognized as a tracer of downward transport from the stratosphere to the troposphere (e.g., Husain et al., 1977; Viezee and Singh, 1980; Sanak et al., 1985; Dibb et al., 1992, 1994; Rehfeld and Heimann, 1995). Husain et al. (1977) reported that pulses of high 7Be concentrations were often associated with air masses of stratospheric origin, as indicated by large potential vorticity. Viezee and Singh (1980) showed that the 7Be concentrations over North America show strong positive correlations with the occurrence of tropopause folding events over several latitude belts. 7Be has also been combined with other radionuclides (e.g., 10Be, 90Sr) as an indicator of transport of stratospheric air to the troposphere (Raisbeck et al., 1981; Rehfeld and Heimann, 1995; Koch and Rind, 1998; Dibb et al., 1994; Jordan et al., 2003; Zanis et al., 2003; Heikkilä et al., 2008a, b). Dutkiewicz and Husain (1985, hereafter referred to as DH85) analyzed 7Be and 90Sr concentrations measured simultaneously in samples from NASA's Global Atmospheric Sampling Program (GASP) and showed that on an annual basis the stratosphere contributed ∼ 25 % of the observed 7Be concentration at the northern mid-latitude surface (∼ 40 % during late spring but only 10 % during fall).

Beryllium-7 is also a useful tracer for vertical mixing and subsidence in the troposphere. Feely et al. (1989) examined the factors that contribute to seasonal variations in 7Be concentrations in surface air. They found that the influences of variations both in the STE rate and in the tropospheric vertical mixing rate are evident in concentrations at most sites in mid-latitudes. Convective transport carries surface air upward and brings down the 7Be at higher altitudes to the surface layer. This is also reflected by the 7Be / 210Pb ratio that peaks at the surface in summer when convective activity is at its maximum (Koch et al., 1996). On the other hand, despite the UT/LS source of 7Be and the continental surface source of 222Rn (precursor of 210Pb), 7Be concentrations have been reported to be positively correlated with 210Pb concentrations, reflecting mixing of subsiding middle- and upper-tropospheric air with continental lower-tropospheric air (Li et al., 2002; Dibb, 2007).

A number of observational studies have demonstrated the feasibility of using 7Be to infer the contribution of ozone-rich stratospheric air to ozone concentrations at ground level (e.g., Husain et al., 1977; Tsutsumi et al., 1998; Helmig et al., 2007) and in the free troposphere (e.g., Johnson and Viezee, 1981; Prospero et al., 1995; Graustein and Turekian, 1996; Kritz et al., 1991; Dibb et al., 2003). These studies are usually based on the correlations between concurrent measurements of ozone and 7Be (as well as other tracers such as water vapor and calculated potential vorticity), with positive 7Be-ozone correlations indicating the presence of the upper-tropospheric or stratospheric air. For instance, Helmig et al. (2007) showed a year-round correlation of ozone with 7Be at Summit, Greenland and concluded that surface-layer photochemical ozone production does not appear to have a noticeable influence on surface ozone levels. However, it is important to note that under some circumstances the observed positive correlations of surface ozone with 7Be may simply reflect the common vertical trends of tropospheric 7Be and ozone and does not necessarily indicate the influence of stratospheric air (Li et al., 2002). Recent global modeling studies showed the models' capability to reproduce the observed 7Be-ozone relationships, providing useful constraints on the stratospheric (vs. photochemical) contribution to tropospheric ozone in the model (Li et al., 2002; Allen et al., 2003; Liu et al., 2004).

Though correct representation of STE is essential for simulating 7Be, ozone and other trace species in the troposphere (e.g., Riese et al., 2012), large variations exist among models. Stevenson et al. (2006) reported the average STE flux of ozone from 26 models of 552 ± 168 Tg year-1. Observation-based estimates of STE fluxes of ozone into the troposphere are typically in the range of 400–600 Tg year-1 (Murphy and Fahey, 1994). Some global models are able to produce STE fluxes of ozone in this range (e.g., Olsen et al., 2004; Hsu et al., 2005; Hsu and Prather, 2009; Lin et al., 2012; Young et al., 2013; Škerlak et al., 2014). For those models with too fast (or rarely, too slow) cross-tropopause transport of ozone, one way to overcome the difficulty is to use the Synoz (synthetic ozone) method (McLinden et al., 2000). The Synoz method involves constraining the global mean cross-tropopause ozone flux to match a prescribed value consistent with observations (e.g., Bey et al., 2001). But this method yields an unrealistic stratospheric ozone field and therefore does not allow for on-line calculations of total ozone columns and photolysis rates and/or heating rates (McLinden et al., 2000). By contrast, the other simple model for stratospheric ozone (linearized ozone or Linoz) developed by McLinden et al. (2000) enables these on-line calculations by linearizing the ozone tendency about the local ozone mixing ratio, temperature, and the overhead column ozone density. Linoz is computationally efficient and can be readily incorporated in climate models for long-term integrations. Nevertheless, using Linoz (or full stratospheric chemistry) in global CTMs or chemistry-climate models that focus on the troposphere requires a realistic model representation of net cross-tropopause total mass fluxes. In this context, 7Be tracer simulations may provide a simple way of evaluating cross-tropopause transport in these models.

The intermodel differences in the estimated intensity and frequency of STE have been attributed to different meteorological fields used to drive the models as well as different transport algorithms and chemistry processes (Cristofanelli et al., 2003). The GMI modeling framework facilitates the reduction of uncertainties of this kind. It is a modular CTM with the ability to incorporate different inputs and components (e.g., meteorological fields, emission inventories, chemical and microphysical mechanisms, and numerical schemes) that represent the different approaches of current models. One of the distinct features of the GMI CTM is the ability to be driven by different meteorological data sets (e.g., Douglass et al., 1999; Considine et al., 2005; Liu et al., 2007) while maintaining the same algorithms for transport, deposition, emission, chemistry and other pertinent processes. This allows us to isolate the uncertainties in the model simulations due to differences in the meteorological data sets alone. The number of factors that may contribute to differences in the simulations is thus reduced, as we previously showed using the GMI simulated 222Rn and 210Pb radionuclide tracers (Considine et al., 2005).

In this paper, we present simulations of atmospheric 7Be distributions with the GMI CTM driven by four different meteorological data sets, including output from GEOS1-STRAT, GISS II′ GCM, fvGCM, and GEOS4-DAS, each featuring significantly different STE characteristics. The reader is referred to Table 1 for a list of acronyms of models and their driving meteorological data sets. We use here not only the meteorological fields that are well known to have reasonably good representations of STE (e.g., fvGCM), but also those with poor representations (e.g., GEOS1-STRAT). The variability in simulated STE allows us to examine and assess the utility of 7Be for evaluating STE in these (and other) global meteorological fields. We will illustrate the consequences of incorrect STE in terms of the simulation of tropospheric 7Be and show that 7Be concentrations and deposition fluxes may be used routinely as a first-order assessment for cross-tropopause transport in global models. We will discuss how the constraints on STE from 7Be are consistent with previous modeling studies of tropospheric ozone using the same meteorological fields.

The remainder of this paper is organized as follows. Section 2 gives a brief description of the GMI model, 7Be source and cross-tropopause flux, and 7Be and ozone observational data sets used for evaluating the model. Section 3 evaluates model results with UT/LS and surface 7Be data. Section 4 assesses cross-tropopause transport of 7Be in different meteorological fields. Section 5 compares the results with previous modeling studies. Section 6 discusses the implications for the impact of STE on tropospheric ozone, followed by the summary and conclusions in Sect. 7.

In this section, we present model results of 7Be simulations driven by four meteorological archives and evaluate them against long-term measurements at the surface and in the UT/LS. Figure 3 shows the annual zonal mean concentrations (in units of millibequerel per standard cubic meter or mBq SCM-1) of 7Be in the four radionuclide simulations using GMI CTM. All four simulations overall show a similar pattern of tropospheric distribution. The highest concentrations are seen in the dry subsiding subtropics. Lowest 7Be concentrations in surface air are found in the Southern Hemisphere mid-latitudes owing to scavenging by frequent large-scale precipitation (Fig. 1). Low 7Be concentrations are also associated with ITCZ, which is characterized by strong convergence and convective precipitation. It appears, however, that among all four simulations the GEOS1-STRAT simulation gives the highest concentrations in the subtropics and the GISS simulation shows the highest concentrations in the high latitudes. This is partly attributed to the differences in the latitudinal distribution of total precipitations in these meteorological archives (Fig. 1).

Figure 4 compares four 7Be simulations in the upper troposphere and/or lower stratosphere (UT/LS) with climatological distributions constructed from the 7Be data contained in the RANDAB database, following Considine et al. (2005) who previously made a similar comparison for 210Pb. Model output are sampled at the months, longitudes, latitudes, and altitudes of the 7Be observations. Figure 4a compares the meridional distribution of 7Be measurements made in the 12–16 km altitude range with the four GMI simulations. Figure 4b shows the same comparison, but for the 16–20 km altitude range. The 12–16 km (about 200–100 hPa) range lies within the upper troposphere in the tropics and the lower stratosphere at mid to high latitudes. The 16–20 km (about 100–50 hPa) range lies within the stratosphere at all latitudes.

At 12–16 km (Fig. 4a), the observations indicate comparatively low tropical upper tropospheric values of ∼ 35 mBq SCM-1, with increasing trends toward high latitudes. The distribution is nearly symmetric about the equator, with more observations available in NH high latitudes. This latitudinal distribution of 7Be concentrations reflects a larger production of 7Be in the lower stratosphere at high latitudes and precipitation scavenging associated with deep convection in the tropics. All four simulations capture the observations at 12–16 km reasonably well. The differences between the four simulated 7Be concentrations are comparable or smaller than the error limits.

At 16–20 km (Fig. 4b), the observations show a tropical minimum of ∼ 150 mBq SCM-1, with increasing concentrations toward high latitudes in both hemispheres. In the tropics and the SH, the four 7Be simulations indicate small differences. In the NH, the four 7Be simulations reveal large differences and bracket the observations. In particular, the GMI/GEOS1-STRAT simulation gives the lowest 7Be concentrations among the four simulations and is lower than the observations. This appears to be due to excessive cross-tropopause transport in GEOS1-STRAT, as further discussed below. On the other hand, as we will also discuss later, the fvGCM and GEOS4-DAS meteorological fields have reasonable cross-tropopause transport. In the latter case, stratospheric 7Be concentrations are primarily determined by a balance between production and radioactive decay in the stratosphere. Therefore the slightly overestimated 7Be at 16–20 km suggests a slightly overestimated global production rate of 7Be in the LP67 source. The Usoskin and Kovaltsov (2008) source, which is about 25 % lower than the LP67 source, would probably yield better agreements with the 7Be observations in the lower stratosphere.

Figure 5a compares the simulated and observed annual average concentrations of 7Be near the surface as a function of latitude. Observed data are from the EML SASP database and are averaged into 10∘ latitude bins. Observations from sites with elevation higher than 500 m are not included because of uncertainties involved in sampling coarse-resolution models at high elevation sites. Model results are sampled at observation locations on a monthly basis. Figure 5b shows the annual zonal mean surface 7Be concentrations in the model to indicate the global representativeness of the averages over the sampling sites. The observations indicate concentration maxima in the subtropics associated with subsidence and minima in the tropics. The tropical minimum reflects rapid scavenging within the ITCZ. Low 7Be concentrations are also observed at mid-latitudes due to efficient scavenging in the mid-latitude storm tracks. Latitudinal trends (i.e., minima and maxima) of 7Be concentrations are well simulated with all meteorological fields except GISS II′. The GMI/GISS simulation shows too high 7Be concentrations at high latitudes; this is because of the well-known excessive cross-tropopause transport at high latitudes in the GISS II′ meteorological fields (e.g., Koch and Rind, 1998; McLinden et al., 2000; Shindell et al., 2003). The overall positive biases in all simulations are partly due to our correction of the long-term records of 7Be observations (by a factor of 0.72) to the 1958 solar maximum source (Sect. 2.5). We find that without this correction, the biases would be significantly reduced.

Figure 5c compares the model-simulated annual mean total deposition fluxes of 7Be at 25 northern mid-latitude sites from which long-term records of observations are available. The 7Be deposition flux observations are from the compilation of Koch et al. (1996), previously used in Liu et al. (2001). The data from individual sites are averaged over 4∘ latitude bins. The model is sampled at observation locations. Figure 5d shows the annual zonal mean total deposition fluxes of 7Be in the model to indicate the global representativeness of the sites. The observations show a maximum (∼ 2100 Bq m-2 yr-1) in the subtropics (∼ 30∘ N) and the fluxes fall off with increasing latitude. The four 7Be simulations show large discrepancies especially in the subtropics (∼ 30∘ N). Overall, the GMI/fvGCM simulation agrees better with the magnitude of the observed fluxes while the GMI/GEOS4 simulation yields better latitudinal trends. GMI/GEOS4 simulates best the observations at the latitudes of 45–60∘ N, but overestimates the observations by ∼ 50 % at 20–40∘ N. The GMI/GISS simulation overestimates the observations at higher latitudes (45–60∘ N) by a factor of ∼ 2. The GMI/GEOS1-STRAT simulation overestimates the observed 7Be deposition fluxes at subtropical latitudes by up to a factor of 2.5 (30∘ N). As with the above model-observation comparison of surface 7Be concentrations, the overall positive biases in model total deposition fluxes would be lower without the correction of 7Be observations (by a factor of 0.72) to the 1958 solar maximum source (Sect. 2.5). However, we will show in the next section that these overestimated 7Be deposition fluxes are largely due to model excessive cross-tropopause transport, especially with the GEOS1-STRAT and GISS II′ meteorological fields.

The above results indicate different levels of success with four meteorological fields in reproducing long-term records of surface and UT/LS 7Be concentrations as well as total deposition fluxes. In this section, we quantify the contribution of 7Be produced in the stratosphere to tropospheric 7Be concentrations and deposition fluxes, followed by an assessment of cross-tropopause transport of 7Be in the meteorological fields used.

Figure 6a shows the stratospheric fraction (%) of annual zonal mean atmospheric 7Be concentrations (i.e., fraction of atmospheric 7Be produced in the stratosphere) in the model simulations as a function of latitude and pressure. The fractions of significantly less than 100 % in the lower stratosphere in all four simulations reflect mainly the seasonal movement of the tropopause. With GEOS1-STRAT, stratospheric contribution to lower-tropospheric 7Be concentrations maximizes at 25–50∘ N (35–45 %) and 25–40∘ S (30–35 %). The tropical middle and upper troposphere show the minimum in stratospheric impact (< 30 %). With GISS II′, the stratospheric contribution to lower-tropospheric 7Be concentrations peaks (30–40 %) at southern high latitudes and remains nearly constant (30–35 %) north of 30∘ N while it is quite small (< ∼ 10–20 %) in the tropical middle and upper troposphere. The strong gradients in the subtropics suggest that the tropics are strongly isolated from the mid-latitudes in the GISS II′ meteorological field. fvGCM and GEOS4-DAS show a similar pattern of stratospheric influence on the troposphere; both indicate maximum contribution from stratosphere near 30–35∘ N (∼ 25 %) and 25–30∘ S (∼ 20–25 %) in the lower troposphere. However, GEOS4-DAS shows larger contributions from the stratosphere to the troposphere (especially the free troposphere) than fvGCM does by a few percent, consistent with the overestimated deposition fluxes at 20–40∘ N by GEOS4-DAS (Fig. 5c). The area of minimal stratospheric influence in the tropics is also narrower in GEOS4-DAS.

Figure 6b shows the stratospheric fraction (%) of annual zonal mean surface 7Be concentrations and that of annual zonal mean 7Be total deposition fluxes (Bq m-2 yr-1) in the model simulations as a function of latitude. With all meteorological fields except GISS II′, maximum stratospheric contribution to total deposition fluxes (vs. surface 7Be concentrations) is shifted toward higher latitudes, reflecting scavenging by frequent mid-latitude precipitation and the dry subsidence in the subtropics. Stratospheric fractions of surface 7Be concentrations at NH mid-latitude are about 38 % (GEOS1-STRAT), 33 % (GISS II′), and 23-24 % (fvGCM and GEOS4-DAS). As discussed in Sect. 2.4, the observed 7Be / 90Sr ratio suggests that 23–27 % of the 7Be in surface air at northern mid-latitudes is of stratospheric origin (DH85). According to this constraint, cross-tropopause transport of 7Be and subsequent transport to the surface in the GEOS1-STRAT and GISS II′ meteorological fields is excessive. On the other hand, it should be noted that the fvGCM and GEOS4-DAS simulations show results remarkably consistent with the DH85 constraint, suggesting that stratospheric influences on surface 7Be concentrations in these two meteorological fields are reasonable. However, DH85 did not provide constraints on latitudinal variation of stratospheric influence on surface 7Be. Of the four meteorological fields, GEOS1-STRAT, fvGCM and GEOS4-DAS show very similar latitudinal distribution of stratospheric influence at the surface (i.e., peak in the subtropics and valley in the tropics or polar regions). By contrast, GISS II′ shows the largest impact of the stratosphere at high latitudes.

Similarly, as shown above, the model overestimates the long-term records of 7Be deposition flux observations at mid-latitudes (and subtropics) with GEOS1-STRAT and at high latitudes with GISS II′ (Fig. 5c). Interestingly, the fvGCM (and to a lesser extent GEOS4-DAS) simulation yields 7Be deposition fluxes close to the observations. This suggests that the DH85 constraint and observed 7Be deposition fluxes are two complementary constraints on cross-tropopause transport of 7Be. We therefore use the DH85 constraint to assess the cross-tropopause transport of 7Be in the meteorological fields.

Using the approach described in Sect. 2.4 (i.e., reduced cross-tropopause transport flux by artificially scaling down the stratospheric 7Be source in the simulation of tropospheric 7Be), we determine the scaling factors for GEOS1-STRAT and GISS to be 1.92 and 1.35, respectively. With the adjustment of 7Be cross-tropopause fluxes for GEOS1-STRAT and GISS, the model calculated stratospheric fraction of 7Be concentrations in surface air at NH mid-latitudes is indeed close to 25 % (i.e., agree with the DH85 constraint) (Fig. 7), thus supporting the validity of Eq. (5). With the adjustment, some simulations also simulate better surface 7Be concentrations and total deposition fluxes at the subtropics (GEOS1-STRAT) and at high latitudes (GISS II′) (Fig. 8 vs. Fig. 5). The improvement is clearer for total deposition fluxes than for surface concentrations. As discussed below, on a global scale total deposition fluxes are sensitive to STE fluxes of 7Be into the troposphere, while surface concentrations are principally dependent on the overall wet removal rate.

Table 3 shows the annual average global budgets of tropospheric 7Be in the four GMI simulations. With an adjustment of 7Be cross-tropopause fluxes, the global burdens and residence times of tropospheric 7Be in GMI/GEOS1-STRAT and GMI/GISS are reduced. In GMI/GEOS1-STRAT the source and sink terms become much closer to that in fvGCM and GEOS4-DAS. A reduction of global 7Be STE fluxes of 0.04 g day-1 results in a decrease of total deposition fluxes of 0.03 g day-1 and radioactive decay of 0.01 g day-1. In GMI/GISS the changes in the budget terms are relatively small due to the smaller adjustment of 7Be cross-tropopause fluxes. Nevertheless, a reduction of global 7Be STE fluxes of 0.01 g day-1 results in a decrease of total deposition fluxes of 0.01 g day-1. These calculations indicate that globally the 7Be total deposition fluxes are sensitive to STE fluxes of 7Be into the troposphere.

The model calculated stratospheric fraction of 7Be in the troposphere may be sensitive to the model diagnosed location of the tropopause, for which there is some uncertainty. For instance, Stajner et al. (2008) used four different definitions of the tropopause on the basis of temperature lapse rate (World Meteorological Organization or WMO definition), potential vorticity (PV), and isentropic surfaces or ozone surfaces. They found that the WMO tropopause was about 0.7–1 km (in the northern mid-latitude) or 0.5–1 km (in the tropics) higher than the ozone or PV determined tropopause. We examine the sensitivity of model diagnosed stratospheric fraction of tropospheric 7Be concentrations to the location of tropopause (not shown) by lowering tropopause height by one model level (approximately 1.2, 1.7, 1.1, and 1.1 km for GEOS1-STRAT, GISS II′, fvGCM and GEOS4-DAS, respectively). Results indicate that stratospheric fractions of surface 7Be concentrations increase by 5–10 %, thus requiring larger adjustments of cross-tropopause transport of 7Be in the meteorological fields in order to meet the DH85 constraint. This also suggests that using the DH85 constraint requires relatively high vertical resolution near tropopause in the model.

While the model diagnosed stratospheric fraction of tropospheric 7Be concentrations is mainly determined by the STE processes in the UT/LS, it may also be sensitive to precipitation scavenging and convective transport in the troposphere. Figure 9 shows the latitude-pressure cross sections of the differences in the stratospheric fraction (%) of annual zonal mean tropospheric 7Be concentrations between the standard simulation and a simulation where precipitation scavenging is turned off. Also shown are the corresponding differences near the surface. The stratospheric fraction of tropospheric 7Be is found to be only weakly dependent on precipitation scavenging, with < 5 % change in most of the troposphere and < 2.5 % change near the mid-latitude surface. Figure 10 shows a similar plot, except that convective transport and scavenging are turned off in the sensitivity simulation. Similarly, the stratospheric fraction of tropospheric 7Be is not sensitive to convective transport and scavenging processes, with < 1 % changes near the mid-latitude surface.

In this section we compare the GMI CTM results for cross-tropopause transport of 7Be with previous modeling studies based on the same or similar meteorological fields.

Liu et al. (2001) found that STE flux of 7Be was overestimated with the GEOS1-STRAT fields in the GEOS-Chem model, consistent with this study using GMI CTM. However, Liu et al. (2001) found that the reduction required to match the DH85 constraint is a factor of 3.5 for the GEOS1-STRAT archive with 4∘ × 5∘ resolution, compared to a factor of 1.92 in the present study. The larger reduction in the former reflects the inclusion of ice particle gravitational settling effect, which results in increased transport from the upper to lower troposphere, as well as the inclusion of the diagnosed tropopause model layer as part of the stratosphere (vs. the troposphere). Interestingly, when specifying ozone concentrations in the lower stratosphere (70 hPa) and letting the model (GEOS-Chem) transport this ozone as an inert tracer into the troposphere, Bey et al. (2001) found a similar overestimate in an ozone simulation with the GEOS-1 data, as diagnosed by the simulation of tropospheric ozone concentrations at high latitudes in winter where transport from the stratosphere is a major source. This indicates that the simulation's deficiency in cross-tropopause transport as diagnosed using 7Be tracers has similar consequences for cross-tropopause transport of ozone.

Koch and Rind (1998) used a 31-layer version of the GISS GCM to simulate 7Be and 10Be and used tropospheric 10Be / 7Be as indicator of STE. Based on limited observations, they suggested that leakage into the troposphere is somewhat excessive in the model, particularly at high latitudes. Using the GISS II′ GCM, McLinden et al. (2000) found that a large fraction of the cross-tropopause transport of ozone occurs at the poles which is inconsistent with the current understanding of stratosphere–troposphere exchange, despite the fact that the global stratosphere–troposphere exchange fluxes of ozone compare well with their best estimate of 475 ± 120 Tg year-1 based on measurements and tracer-tracer correlation. Shindell et al. (2003) presented an updated version of the GISS II′ climate model which still overestimates ozone in the middle troposphere at high latitudes, likely reflecting deficiencies in the model's downward transport of stratospheric air. Our conclusions about cross-tropopause transport of 7Be in GISS II′ in this work are consistent with these previous studies. Overestimated STE fluxes of 7Be as diagnosed in GMI/GISS based on the DH85 constraint simply reflect the incorrect latitudinal distribution of cross-tropopause transport, that is, too fast STE at higher latitudes and too slow STE at lower latitudes. The DH85 constraint was only applicable and applied for NH mid-latitude surface and thus does not provide constraint on the model global STE flux of 7Be if the latitudinal distribution of STE is incorrect.

The large-scale stratospheric transport (Brewer–Dobson circulation) in fvGCM has been shown to be realistic (Douglass et al., 2003) and mean age of stratospheric air is similar to observations (Strahan and Douglass, 2004; Douglass et al., 2008; Strahan et al., 2009). This suggests credible cross-tropopause transport of mass and ozone in fvGCM because the large-scale exchange between the stratosphere and troposphere is largely tied to the Brewer–Dobson circulation through the overworld wave driving (Holton et al., 1995; Olsen et al., 2004). Based on this finding, the meteorological data from fvGCM were used to drive GMI CTM by several authors to study tropospheric ozone. Considine et al. (2008) evaluated near-tropopause ozone distributions with ozonesonde data. Terao et al. (2008) examined the role of variability in the input of stratospheric ozone on the interannual variability of tropospheric ozone in the northern extratropics. Liang et al. (2009) investigated the impact of stratosphere-to-troposphere transport on tropospheric ozone and NOx chemistry over the Arctic. By contrast, GEOS4-DAS tends to have too strong a residual circulation, and the age of air is too young compared to observations (Schoeberl et al., 2003; Schoeberl, 2004; Douglass et al., 2008). A GMI CTM simulation driven with the GEOS4-DAS meteorological fields showed the model's inadequacy in simulating upper-tropospheric ozone (Liang et al., 2009). These findings are consistent with what we illustrated in this study from a perspective of 7Be tracers. That is, GEOS4-DAS features a larger impact of STE on the troposphere (especially UT) than fvGCM does, while the latter has more credible cross-tropopause transport as constrained by observed 7Be deposition fluxes (Fig. 5c) and the DH85 criterion (Fig. 6).

In this section we discuss the implications of different characteristics of cross-tropopause transport of 7Be for stratospheric influence on tropospheric ozone in different meteorological fields. At the time of this study, the GMI full-chemistry model can be driven with GEOS1-STRAT, fvGCM and GEOS4-DAS, but not GISS II′ meteorological fields. This allows us to examine any potential relationship between the cross-tropopause transport of 7Be and ozone when these fields are used to drive the model.

Ozonesonde, surface and satellite observations provide useful constraints on the stratospheric contribution to tropospheric ozone (e.g., Rind et al., 2007; Lin et al., 2012). Figure 11 shows comparisons of model tropospheric ozone profiles with annual mean ozonesonde observations for a range of latitudes (Considine et al., 2008). These results are typical of other stations at similar latitudes. The GMI/GEOS1-STRAT simulation produces excessive ozone throughout the troposphere at all latitudes except in the tropics while the GMI/fvGCM and GMI/GEOS4-DAS simulations are generally in agreement with the observations (with slightly overpredicted ozone in the mid-latitude upper troposphere). The GEOS1-STRAT simulation has the largest overestimate of O3 in spring. We also compared model surface ozone concentrations with the Logan (1999) surface ozone data set (not shown). Among the three GMI simulations, the GMI/GEOS1-STRAT simulation shows the largest errors in surface ozone concentrations during winter and spring when stratospheric contribution is at its peak. These are in line with the relative magnitudes of cross-tropopause transport efficiencies of 7Be in the three meteorological fields (i.e., too fast STE in GEOS1-STRAT), discussed in previous sections. Indeed, the tropospheric version of the GMI/GEOS1-STRAT model with constrained STE flux of ozone using the Synoz approach (about 579 Tg year-1) simulates ozonesonde observations of tropospheric ozone reasonably well (dotted line, Fig. 11).

Figure 12 shows GMI simulated annual zonal mean tropospheric ozone column (TOC), in Dobson Units, compared with observed climatologies from TOMS/SBUV (1979–2005; Fishman et al., 2003) and OMI/MLS (October 2004–July 2008; Ziemke et al., 2006). The WMO definition of thermal tropopause is used to calculate the model TOC. While the GMI/fvGCM and GMI/GEOS4-DAS simulations are similar and overestimate TOC by up to ∼ 20 DU, the GMI/GEOS1-STRAT simulation overestimates TOC by as much as ∼ 40 DU. The excessive O3 in the GMI/GEOS1-STRAT simulation with maxima at 30∘ N and 30∘ S suggests downward transport of ozone from the stratosphere is too fast. The tropospheric version of the GMI/GEOS1-STRAT model with constrained STE flux of ozone provides a much better simulation of global TOCs (red dashed line, Fig. 12), which are comparable to those from GMI/fvGCM and GMI/GEOS4-DAS simulations. However, model TOCs are still ∼ 10–14 DU larger than satellite observations in the subtropics and mid-latitudes. Previously, Ziemke et al. (2006) considered uncertainties in both model and observations and subjectively interpreted model-OMI/MLS TOC differences of 10 DU and higher as being significant. As Stajner et al. (2008) noted, a low extratropical tropopause used by Ziemke et al. (2006) may have played an important role in the underestimation of OMI/MLS TOC. Yang et al. (2010) also found that their OMI/MLS potential vorticity mapped TOCs are smaller than ozonesonde TOCs by 5.9 DU with a standard deviation of the differences of 8.4 DU. On the other hand, the GMI/fvGCM simulation tends to overestimate ozone just below the tropopause at mid-latitudes (Fig. 11); these biases do not appear to be due to excessive stratospheric influence (Considine et al., 2008). Current global models also tend to overpredict surface ozone during summer and early fall over the eastern US and Japan (Fiore et al., 2009). Therefore the simulated TOCs are very likely biased high.

We further examine the relationship between the cross-tropopause transport of 7Be and ozone with the GEOS1-STRAT meteorological fields, in which case STE is known to be too fast. Figure 13a shows the latitudinal variations of annual zonal mean tropospheric 7Be column overestimate (Δ7Be) and TOC overestimate (ΔTOC) in the GMI/GEOS1-STRAT simulation. Δ7Be is obtained by subtraction of the STE-flux-adjusted simulation (Sect. 2.4) from the standard simulation. ΔTOC is obtained by subtracting the GMI tropospheric model simulation (with STE flux of ozone about 579 Tg year-1) from the GMI full-chemistry model simulation. Figure 13b shows the correlation between the global distributions of Δ7Be and ΔTOC. The lines of best fit are calculated using the reduced-major-axis (RMA) method (Hirsch and Gilroy, 1984). Standard errors for the intercept and the slope are computed as described by Miller and Kahn (1962). Overall, the location of overestimated ozone follows that of overestimated 7Be, with both maxima near 30∘ N and 30∘ S. The strong correlation between Δ7Be and ΔTOC implies that 7Be is a good indicator of cross-tropopause transport of ozone. These support our conclusion that 7Be is a useful utility for assessing cross-tropopause transport of ozone in global models.

We have assessed the ability of the Global Modeling Initiative (GMI) chemical transport model (CTM) using different meteorological data sets to simulate the atmospheric distributions of 7Be, a natural aerosol tracer originating from the upper troposphere and/or lower stratosphere and removed from the troposphere primarily by wet deposition. The model was driven by four meteorological data sets (GEOS1-STRAT, GISS II′, fvGCM, GEOS4-DAS) which feature significantly different cross-tropopause transport characteristics. The GMI modeling framework was configured such that the variability between the simulations mainly reflects the use of different meteorological data. Our goal was to assess the utility of 7Be as a tracer of cross-tropopause transport in global models and develop a methodology to exploit such a utility. We have also discussed the implications of excessive cross-tropopause transport as revealed by 7Be simulations for the modeling of tropospheric ozone.

We evaluated the four simulations of 7Be with RANDAB, a unique database of upper atmosphere radionuclide climatological observations compiled by the DOE (now DHS) Environmental Measurement Laboratory, as well as long-term measurements at the surface. Model simulations capture the UT/LS observations with respect to latitudes. The GMI/GEOS1-STRAT simulation shows the lowest 7Be concentrations among the four simulations in the lower stratosphere, and underestimates the observations. This reflects the well-known highly overestimated cross-tropopause transport in GEOS1-STRAT DAS. At the surface, GMI/GISS II′ reproduces the observed latitudinal trends of 7Be concentrations, but shows too high concentrations at high latitudes. The GMI/fvGCM simulated 7Be deposition fluxes are the closest to the observations, while the GMI/GEOS1-STRAT overestimates the observed 7Be deposition fluxes at subtropical latitudes by up to a factor of 2.5 (30∘ N) and the GMI/GISS simulations at high latitudes (45–60∘ N) are a factor of 2 too high. We were able to show that the observed 7Be deposition fluxes offer a strong constraint on stratosphere-to-troposphere transport in global models.

We examined the observational constraint from Dutkiewicz and Husain (1985) (DH85) on the stratospheric contribution to tropospheric 7Be using the GMI modeling framework. DH85 analyzed the observed 7Be / 90Sr ratio, which suggests that 23–27 % of the 7Be in surface air at northern mid-latitudes is of stratospheric origin. This constraint offers a sensitive test of cross-tropopause transport in global models. Comparison of the fraction of surface air of stratospheric origin estimated from the 7Be simulations with the DH85 constraint indicates excessive cross-tropopause transport at mid-latitudes with the GEOS1-STRAT meteorological fields and at high latitudes with the GISS II′ fields. Interestingly, these simulations also overestimate observed 7Be deposition fluxes at middle and high latitudes, respectively. With a correction to cross-tropopause flux, the model simulates better surface 7Be concentrations and total deposition fluxes. By contrast, the fvGCM meteorological data yield the most reasonable cross-tropopause transport of 7Be according to the DH85 constraint, consistent with the fact that the GMI/fvGCM simulated 7Be deposition fluxes are closest to the observations. These results illustrate that the GMI framework is very useful for characterizing and helping to reduce uncertainties in the processes such as cross-tropopause transport in the meteorological fields that are used to drive chemical transport models. Note that since wet deposition removes both the stratospheric and tropospheric components of 7Be nondiscriminatively, the model diagnosed fraction of 7Be of stratospheric origin does not significantly depend on the rate of wet removal.

The model diagnosed stratospheric fraction of 7Be in surface air is sensitive to the diagnosed location of tropopause, in particular when the model vertical resolution is relatively coarse (> 1–1.5 km) near the tropopause region. This suggests that stratospheric fraction of 7Be is a more useful diagnostic when the model has sufficient vertical resolution (< 1–1.5 km) so that the tropopause can be well defined. We used the WMO definition of thermal tropopause and include the diagnosed tropopause model layer as part of the troposphere (vs. the stratosphere). As such our assessment of cross-tropopause transport of 7Be in the four meteorological data sets in the GMI CTM is consistent with previous modeling studies of stratospheric influence on tropospheric ozone.

Incorrect cross-tropopause transport of 7Be implies misrepresented downward influx of stratospheric ozone to the troposphere in a model. We demonstrated this by examining the relationship between the cross-tropopause transport of 7Be and ozone as simulated by GMI CTM driven with GEOS1-STRAT, fvGCM and GEOS4-DAS meteorological fields. We found that excessive cross-tropopause transport of 7Be corresponds to overestimated stratospheric contribution to tropospheric ozone, as constrained by ozonesonde, surface and satellite observations.

In summary, the 7Be simulation, which is computationally cheap and technically simple, in combination with the DH85 7Be observational constraint and observed 7Be deposition fluxes may be used routinely to assess cross-tropopause transport in global models. We recommend separate transport of the 7Be produced in the stratosphere (7Be-strat) to evaluate the ratio of 7Be-strat to total 7Be (i.e., beryllium-7 produced in both the stratosphere and the troposphere) in surface air against the DH85 constraint. This can serve as a first-order assessment of cross-tropopause transport in the meteorological fields. With improved estimates of 7Be production rates as well as their year-to-year variations, model multi-year 7Be simulations together with long-term observations would provide useful constraints on the interannual variability of STE. While this study uses 7Be alone, future modeling work will include using 10Be / 7Be, a more sensitive indicator of STE (Rehfeld and Heimann, 1995; Koch and Rind, 1998; Jordan et al., 2003).