GEOSCCM integrates the chemistry mechanism of the Global Modeling Initiative (GMI) chemistry and transport model (CTM; Duncan et al., 2007b; Strahan et al., 2007) into the GEOS-5 Atmospheric GCM (AGCM). The GMI chemistry includes a comprehensive mechanism of tropospheric and stratospheric chemistry, including 117 species and over 400 reactions. The reference simulation for this study, hereafter called RefGMI, is the year 2000 time slice simulation conducted for the ACCMIP intercomparison. The configuration of the ACCMIP simulations is described in Lamarque et al. (2013). The biases in CO and methane lifetime seen in the GEOSCCM simulation are similar to those seen in the ACCMIP multi-model mean (Naik et al., 2013). Consequently, our analyses of bias in the GEOSCCM are likely applicable to other CCMs as well. We use the GMI chemistry option to quantify the impact of changes in CO emissions on methane, OH, and ozone.

GEOSCCM includes a CO-only option to “tag” CO according to source type or location (Bian et al., 2010). This simplified chemistry option allows us to separate the contributions of different CO sources and to quantify the impact of a specific change in OH. In this chemistry option, the loss of CO is calculated based on prescribed OH fields, so changes in CO do not feed back onto OH. Our reference tagged-CO simulation, called RefCOonly, uses monthly OH fields archived from the RefGMI simulation, and the CO sources from methane and isoprene oxidation are calculated using monthly methane and isoprene fields archived from RefGMI as well. Emissions and other forcings are chosen to parallel the RefGMI simulation; however, the CO-only option includes an amplification factor for anthropogenic and biomass burning CO emissions to account for the absence of co-emitted NMHCs (Duncan et al., 2007a). We use the CO-only option to calculate the influence of specific sources on CO concentrations, and to isolate the impact of specific changes in OH on CO.

The GEOSCCM includes an option to constrain the meteorology with MERRA or any GMAO assimilation product. The simulation is pulled towards the MERRA analysis through application of an incremental analysis update (Bloom et al., 1996), calculated every 6 hours from comparison of the simulation with the analysis. We conduct a CO-only simulation with specified dynamics from MERRA, which we refer to as COonlySD, where “SD” stands for “specified dynamics”. The COonlySD simulation has the same emissions and OH field as RefCOonly, but the tracer transport will differ between the two simulations.

A third chemistry option within the GEOSCCM is the CO–OH option (Duncan et al., 2000, 2007a). This chemistry option is of intermediate complexity between the GMI and tagged-CO options. This option is similar to the CO-only option except that OH is calculated interactively, providing a feedback between CO and OH concentrations. In our reference simulation with this option, RefCO–OH, chemistry inputs to the parameterization of OH such as ozone, NOx, CH4, and NMHC concentrations are prescribed from the archived monthly output of the RefGMI simulation. The temperature, water vapor, and irradiance-related variables input to the parameterization are calculated within the CCM. Since the parameterization was only designed to provide tropospheric OH values, we use the results of the parameterization from the surface to 300 hPa, and archived OH fields taken from the RefGMI simulation above 300 hPa. The CO emission amplification factors are adjusted slightly downward compared to the CO-only simulation to bring the RefCO–OH simulation into better agreement with the RefGMI simulation. We use the CO–OH option to examine how biases in a particular model field such as ozone affect OH and CO concentrations.

Adjusting the strength of midlatitude CO sources can reduce the bias in the interhemispheric CO gradient. We use CO-only simulations to examine the impact of increasing specific sources of CO. Following the method of Strode and Pawson (2013), we estimate the impact of increasing a particular source by increasing the tagged-CO tracer for that source and then re-computing total CO. We impose increases of 10, 20, 50, 100, and 150 % for each tagged tracer. Figure 3 shows how the increased Asian anthropogenic CO (COaa) or tropical biomass burning CO (COtrbb) alters the comparison between modeled CO and the GMD CO observations as a function of latitude taken as an average of March to August. Similar results are present for spring and summer individually. An increase in COaa of approximately 100 % removes the negative bias at high-latitude sites but has little effect on the small negative bias at low latitudes (Fig. 3a). CO transported southward from Asia encounters higher OH than that transported northward, leading to a shorter lifetime (Duncan and Logan, 2008). Increasing COtrbb improves the model agreement with observations at low latitudes but creates overestimates at some tropical sites while providing only a modest reduction in the high-latitude bias (Fig. 3b).

We examine the impact of reducing OH concentrations globally or only in the NH in our CO-only simulation and find that a large decrease in NH OH is effective in reducing the high-latitude CO bias. Naik et al. (2013) found that the ACCMIP multi-model mean underestimated annual mean tropospheric OH by 5–10 % globally. The NH / SH OH ratios in the ACCMIP multi-model mean and GEOSCCM ACCMIP simulation are 1.28 and 1.18, respectively (Naik et al., 2013), compared to the Patra et al. (2014) estimate of 0.97 ± 0.12. Figure 3c shows that reducing OH throughout the year by 5–10 % leads to a small increase in CO across all latitudes, yielding a small reduction in the bias compared to GMD observations. Decreasing OH by 20 % in the NH only leads to a large improvement in high-latitude CO compared to observations, with only a small effect in the Southern Hemisphere.

We find that achieving zero bias in both the inter-hemispheric gradient (IHG) and the global mean would require changes in multiple emission sources. Figure 4 illustrates how changing the concentrations of several tagged-CO tracers, as well as OH, impacts the global mean bias, IHG bias, and correlation (r2) of the simulated CO compared to the GMD observations for March through May. Increasing anthropogenic CO from Asia (COaa), Europe (COea), or North America (COnaa), or increasing Russian biomass burning (COrubb) reduces the bias in the IHG as well as in the global mean. Increasing CO from tropical biomass burning (COtrbb) leads to a smaller reduction in the IHG bias, while increasing CO from biogenic emissions (CObio) reduces the mean bias with little effect on the IHG (Fig. 4a). Consequently, increases in both groups of emissions are necessary to simultaneously remove the global mean bias and correct the IHG.

A similar analysis for June through August (Fig. 5) shows that increasing COaa, COea, COnaa, or COrubb can eliminate the majority of the bias in both the global mean and the IHG in summer. Increasing COrubb is more effective in summer than spring since boreal biomass burning emissions are larger in summer. In contrast to the spring results, increasing COtrbb in summer leads to greater bias in the IHG. This difference occurs because tropical biomass burning occurs primarily north of the Equator in March and April but shifts to the Southern Hemisphere in June, July, and August.

While increases in COaa, COea, COnaa, and COrubb show similar slopes for IHG versus global mean bias, increasing COnaa reduces the correlation with observations, whereas increasing the other sources yields a slight improvement in correlation (Figs. 4b, 5b). Increases in both COtrbb and CObio reduce the correlation with observations in summer, but the effect is also present for CObio in spring. We therefore exclude increases in COnaa and CObio in the rest of our study.

Figures 4 and 5 also show the impact of changing OH. The sensitivity of CO to changes in OH is location dependent, with higher sensitivity in regions without strong local CO sources (Holloway et al., 2000). Reducing OH globally reduces both the IHG and global mean bias. However, reducing OH by 20 % in the NH only yields a greater reduction in both biases as well as the greatest improvement in correlation. We refer to the simulation with the 20 % decrease in NH OH as COonlyLowNHOH.

We next compare the impact of increasing emissions with that of reducing NH OH by 20 %. We compare the COonlyLowNHOH scenario with a simulation called GMI-HiEmis that includes increased CO emissions, further described in Sect. 3.3. The RefCOonly simulation shows a similar surface CO distribution to RefGMI (Fig. 6a, c). The COonlyLowNHOH scenario (Fig. 6d) improves the agreement with the remote high-latitude sites compared to the RefCOonly simulation, but like the HiEmis case (Fig. 6b) it leads to an overestimate of CO concentrations over Europe and the eastern United States. Consequently, a combination such as reduced NH OH and reduced emissions over the eastern USA is likely needed to reconcile the simulated CO with observations.

Sparse surface data makes it difficult to determine from comparison with GMD observations whether the higher CO seen in GMI-HiEmis and COonlyLowNHOH is realistic (Fig. 6), so we also compare the four simulations shown in Fig. 6 to the CO column from MOPITT. Both the RefGMI and RefCOonly simulations show a large negative bias in NH CO compared to MOPITT (Fig. 7a, c). The GMI-HiEmis and COonlyLowNHOH simulations both reduce this negative bias (Fig. 7b, d), but the increased emissions of the GMI-HiEmis simulation lead to a greater overestimate of CO over east Asia and Indonesia. Furthermore, eliminating the increase in COea to reduce the high bias compared to surface observations over Europe and compensating with a larger adjustment in Asian CO would lead to an even greater bias over Asia compared to MOPITT.

Transport, in addition to chemistry and emissions, plays a role in the IHG of CO. The simulations discussed so far are free-running CCM simulations driven by SSTs, since our goal is to understand the biases seen in CCM studies such as ACCMIP. Consequently, the simulated tracer transport is affected by any differences between the simulated and actual meteorology. We examine the sensitivity of the CO bias to model transport by comparing RefCOonly, which is a free-running CCM simulation, with the COonlySD simulation, which has year-specific meteorology from MERRA. The largest difference between the simulations occurs poleward of 30∘ N during July through October (Fig. 8a). The use of specified meteorology makes only a small difference in the global mean and IHG CO biases compared to surface observations in March–May (Fig. 4), but it leads to a large reduction in bias as well as improved spatial correlation with the GMD observations in June–August (Fig. 5). The biases in global mean CO and the CO IHG are -3.5 and -6.6 %, respectively, in COonlySD, compared to -11 and -26 %, respectively, in RefCOonly. More CO from Northern Hemisphere anthropogenic and boreal biomass burning sources remains in the lower troposphere and reaches the high latitudes in COonlySD, whereas more is transported to the upper troposphere in RefCOonly (Fig. 8b).

Comparison of output from the CCMs that participated in the ACCMIP study to OMI/MLS TCO reveals positive biases in simulated tropospheric ozone over the NH midlatitudes and negative biases over the tropical Pacific and SH (Young et al., 2013), which lead to an overestimate in NH OH production and an underestimate in SH OH production (Naik et al., 2013). The multi-model mean tropospheric ozone also shows a high bias in the NH and low bias in the SH compared to the Tropospheric Emission Spectrometer (TES) (Bowman et al., 2013). A multi-species assimilation study by Miyazaki et al. (2012b) found that assimilating TES ozone increased OH concentrations in the SH.

We use the CO–OH option to investigate the impact of removing the GEOSCCM's tropospheric ozone column bias relative to the OMI/MLS observations (Fig. 9). We scale the tropospheric ozone values input to the parameterization of OH for each month between October 2004 and December 2010 from 60∘ S to 60∘ N so that the tropospheric ozone column is unbiased compared to the OMI/MLS TCO for that month. No scaling is applied where the TCO data are missing. We use the scaled ozone input in a sensitivity study for October 2004 to January 2010, called CO–OHSensTCO, which parallels the RefCO–OH simulation but uses the scaled tropospheric ozone values in the parameterization of OH. CO–OHSensTCO shows a small (2 %) decrease in global mean OH compared to RefCO–OH, with a 3 % decrease in NH OH and 1 % decrease in SH OH. The increased OH leads to a 2 % increase in both methane and methyl chloroform lifetimes against OH. The NH / SH OH ratio decreases slightly to 1.19 (Table 4), suggesting that the IHG in the tropospheric ozone bias makes only a minor contribution to the model's interhemispheric OH asymmetry.

We next examine how biases in stratospheric ozone affect OH through their role in photolysis. Voulgarakis et al. (2013) found that changes in stratospheric ozone and tropospheric OH in the ACCMIP models both correlated strongly with J(O1D). We conduct a sensitivity study for 2005 through 2009, CO–OHSensStO3, which parallels the RefCO–OH simulation but replaces the simulated ozone with the GMAO ozone assimilation in the stratosphere for input into the parameterization. The change in global mean OH in CO–OHSensStO3 is nearly identical to that of CO–OHSensTCO, but CO–OHSensStO3 places more of the change in the SH, causing the NH / SH OH ratio to increase to 1.23 (Table 4). The change in OH due to stratospheric ozone bias is small in part because GEOSCCM has relatively small biases in tropical stratospheric ozone. However, the multi-model Chemistry Climate Model Validation-2 (CCMVal-2) study (Morgenstern et al., 2010) shows a large spread across models for column ozone in the tropics (Austin et al., 2010). Thus, biases in stratospheric ozone may play a larger role in tropospheric OH in models with larger stratospheric ozone biases.

We investigate the impact of model biases in water vapor on OH concentrations through its role in OH production. Inter-model differences in OH in the POLMIP study are correlated with inter-model differences in simulated water vapor (Monks et al., 2015). The GEOS-5 AGCM exhibits a high bias in specific humidity compared to the MERRA reanalysis in much of the troposphere (Molod et al., 2012), and a high bias in the midtroposphere is also seen throughout the year in GEOSCCM compared to Atmospheric Infrared Sounder (AIRS) data (Lamarque et al., 2013).

We quantify the impact of this bias by conducting a sensitivity study, CO–OHSensQ, that applies altitude- and latitude-dependent zonal mean scaling factors for each month to the specific humidity provided to the OH parameterization. The scaling factors are based on comparison of the RefGMI simulation to MERRA and are applied in 100 hPa intervals between 900 and 200 hPa for 60∘ S to 60∘ N. These scaling factors are designed to remove the mean bias in the simulated specific humidity compared to MERRA while allowing the simulated water vapor to vary in space and time in a manner consistent with the simulated meteorology. The water vapor scaling is applied only to the OH parameterization and does not impact the general circulation of the simulation. The scaling results in lower specific humidity in the middle and upper troposphere, with the largest percent reductions occurring between 300 and 500 hPa (Fig. 10).

Simulated OH is 6 % lower in the CO–OHSensQ simulation than the RefCO–OH simulation, and the methane and methyl chloroform lifetimes against OH are thus 6 % longer (Table 4). Consequently, water vapor bias has a larger impact on OH concentrations than the stratospheric or tropospheric ozone biases in our simulations. The OH reduction is similar in both hemispheres in the annual mean, so the NH / SH OH gradient is unchanged compare to RefCO–OH.

Anthropogenic NOx emissions, which contribute to secondary production of OH, are located primarily in the Northern Hemisphere. Consequently, an overestimate of these emissions could contribute to the NH–SH asymmetry in simulated OH. Previous studies have estimated NOx emissions based on satellite observations of NO2 columns (Lamsal et al., 2011; Miyazaki et al., 2012a), but some uncertainty in NOx emissions remains. We quantify the sensitivity of simulated OH to NH NOx with a sensitivity study, CO–OHSensNOx, which parallels RefCO–OH but includes 30 % lower NO and NO2 concentrations in the NH. We note that in this experiment ozone values remain the same as in RefCO–OH, rather than responding to the change in NOx. The reduction in NH NOx emissions leads to a 3 % reduction in global OH (Table 4). SH and NH OH are reduced by 0.3 % and 6 %, respectively, reducing the NH / SH OH ratio to 1.15. While this shows that the hemispheric asymmetry in OH is affected by NOx emissions, a much larger redistribution of NOx emissions would be required to completely eliminate the asymmetry.

In the previous sections, we examined the sensitivity of OH and its IHG to model biases in tropospheric and stratospheric ozone, water vapor, and NH NOx emissions. We find that water vapor has the largest impact on global mean OH, while NH NOx emissions have the largest impact on the NH / SH ratio. However, none of these biases individually explains the 20 % reduction in NH OH that would remove the interhemispheric asymmetry and which Figs. 4 and 5 suggest are necessary to remove most of the CO bias. We note, however, that the CO–OH option simulations do not account for all the chemical feedbacks between ozone, methane, OH, and other species, and they may underestimate the sensitivity of the full chemistry simulation to some of these biases.

Table 4 suggests that applying multiple bias corrections simultaneously would bring our simulation into good agreement with the observation-based estimates of global mean OH. We conduct a final sensitivity simulation, called CO–OHSensAll, incorporating the bias corrections for tropospheric and stratospheric ozone, as well as the specific humidity scaling and 30 % reduction in NH NOx emissions. This simulation results in a 13 % reduction in global mean OH, with a 10 % reduction in the SH and 16 % reduction in the NH. The NH / SH OH ratio is reduced to 1.14. The lifetimes of methyl chloroform and methane in this simulation increase by 15 %. Applying this 15 % increase in methane lifetime to the RefGMI methane lifetime against OH of 9.6 years would yield a lifetime of 11 years, consistent with the 11.2 ± 1.3 year estimate of Prather et al. (2012). This shows that removing the main model biases related to OH production would bring simulated global mean OH into good agreement with observation-based estimates, although some NH–SH asymmetry would remain.

The reduction in OH in the sensitivity studies compared to RefCO–OH leads to higher surface concentrations of CO. The bias in the global mean March–August surface CO changes from -4 % in RefCO–OH to 3 % in CO–OHSensAll, while the bias in the IHG changes from -11 to -4 %. Figure 11 shows the latitudinal distribution of the change in CO for each sensitivity simulation versus RefCO–OH. The sum of the changes for CO–OHSensTCO, CO–OHSensStO3, CO–OHSensQ, and CO–OHSensNOx, shown as the dotted line, is similar but slightly smaller than the change for CO–OHSensAll, indicating small nonlinearities in the system. Correcting water vapor (CO–OHSensQ) makes a large contribution to the CO enhancement in both hemispheres, while adjusting NOx emissions (CO–OHSensNOx) and to a lesser extent tropospheric ozone (CO–OHSensTCO) contributes to the larger increase in the Northern Hemisphere.