Corresponding author: P. K. Patra, Research Institute for Global Change/JAMSTEC, 3173-25 Showa-machi, Yokohama 236–0001, Japan. (firstname.lastname@example.org)
 To assess horizontal and vertical transports of methane (CH4) concentrations at different heights within the troposphere, we analyzed simulations by 12 chemistry transport models (CTMs) that participated in the TransCom-CH4 intercomparison experiment. Model results are compared with aircraft measurements at 13 sites in Amazon/Brazil, Mongolia, Pacific Ocean, Siberia/Russia, and United States during the period of 2001–2007. The simulations generally show good agreement with observations for seasonal cycles and vertical gradients. The correlation coefficients of the daily averaged model and observed CH4 time series for the analyzed years are generally larger than 0.5, and the observed seasonal cycle amplitudes are simulated well at most sites, considering the between-model variances. However, larger deviations show up below 2 km for the model-observation differences in vertical profiles at some locations, e.g., at Santarem, Brazil, and in the upper troposphere, e.g., at Surgut, Russia. Vertical gradients and concentrations are underestimated at Southern Great Planes, United States, and Santarem and overestimated at Surgut. Systematic overestimation and underestimation of vertical gradients are mainly attributed to inaccurate emission and only partly to the transport uncertainties. However, large differences in model simulations are found over the regions/seasons of strong convection, which is poorly represented in the models. Overall, the zonal and latitudinal variations in CH4 are controlled by surface emissions below 2.5 km and transport patterns in the middle and upper troposphere. We show that the models with larger vertical gradients, coupled with slower horizontal transport, exhibit greater CH4 interhemispheric gradients in the lower troposphere. These findings have significant implications for the future development of more accurate CTMs with the possibility of reducing biases in estimated surface fluxes by inverse modeling.
If you can't find a tool you're looking for, please click the link at the top of the page to "Go to old article view". Alternatively, view our Knowledge Base articles for additional help. Your feedback is important to us, so please let us know if you have comments or ideas for improvement.
 Methane (CH4) is an important greenhouse gas, contributes to photochemical smog formation under polluted conditions by reaction with hydroxyl radical (OH), and is the main source of water vapor in the stratosphere [Graedel and Crutzen, 1993]. Over the last two decades, CH4 simulations have been conducted using three-dimensional global chemistry transport models (CTMs) [e.g., Fung et al., 1991]. Recently, a model intercomparison experiment (TransCom-CH4) has been organized with the aim to assess the roles of uncertainties in model transport, surface emissions, and OH chemistry for simulating global CH4. The first analysis [Patra et al., 2011a] concentrated on background concentrations at the surface and showed overall consistent results across the participating models. However, that analysis was limited to eight surface sites, where simultaneous measurements of CH4, sulfur hexafluoride (SF6), and methyl chloroform (CH3CCl3) were available for more than 5 years in the period 1990–2007. Patra et al. [2011a] demonstrated that some models consistently underestimate or overestimate the meridional gradients of all the three species depending on the simulated rate of interhemispheric mixing. Based on SF6 concentration gradients, for two northern (Barrow and Mauna Loa) and two southern (Cape Grim and South Pole) hemispheric sites, the interhemispheric exchange time is estimated to be in the range of 1.0–1.8 years for the suites of TransCom-CH4 models.
 How the simulated zonal or meridional gradient varies with altitude for the long-lived species, however, has not been studied in detail yet. Nakazawa et al.  suggested that the meridional gradient in the carbon dioxide (CO2) concentration in the upper troposphere is significantly smaller compared to that near the Earth's surface. Stephens et al.  showed that the CO2 vertical gradients between the former suites of TransCom (Level 3) models, mostly developed pre-2000, varies widely due to the difference in vertical mixing rate. Note here that the interpretation of model transport has become complex using CO2 due to strong seasonal variation in surface fluxes; strong source due to heterotrophic respiration in addition to fossil fuel emissions in the winter, and strong sink dominated by photosynthetic uptake in the summer [e.g., Miyazaki et al., 2008]. The strong diurnal cycle in CO2 fluxes also complicate flux-transport interaction vertically, particularly over the land region [Law et al., 2008]. Because CH4 sources dominate over the soil uptake on the surface in all seasons, and the atmospheric lifetime of CH4 is about 10 years [Patra et al., 2011a], the TransCom model intercomparison for CH4 vertical profiles would be relatively straightforward for interpreting the transport and emission model errors. Advantages of using constituents (e.g., CH4) with net sources on the Earth's surface over that of CO2 to diagnose model transport have been shown over the South Asian region during the season of strong CO2 uptake by the terrestrial ecosystem [Patra et al., 2011b].
 Analyses of CH4 vertical profiles using CTMs have been limited to single model [Bergamaschi et al., 2009] and for specific regions [Beck et al., 2012] and are proven to be useful for validating surface emissions. The TransCom-CH4 data base includes simulated vertical profiles by 17 CTMs, which can be compared to data from several aircraft measurement programs that have been carried out during the last decade [Tans et al., 1996; Machida et al., 2001; Gatti et al., 2010]. The latitude-pressure distributions of zonal mean simulation results show that CH4 concentrations, within the troposphere, decrease with altitude in the Northern Hemisphere (NH) and increase with altitude in the Southern Hemisphere (SH) [Patra et al., 2011a]. Both the vertical and meridional gradients exhibit seasonal dependency in relation to seasonality in transport, surface emissions, or chemical loss. However, this conjecture is based only on the model intercomparison, without comparison with observations. Thus, an evaluation of the modeling results with aircraft observations will likely improve our understanding of CH4 transport mechanisms of vertical and interhemispheric exchanges in the troposphere.
 The purpose of this study is to evaluate the simulated CH4 concentrations using aircraft measurements. The model-observation difference depends on uncertainties in surface fluxes, model transport, and chemistry. We assess the flux and transport dependence of the model-observation differences using vertical profiles over a variety of surface conditions and transport regimes, e.g., wetlands in Siberia, tropical rainforests in Amazon, boreal forests of North America, and the marine environment. In the next sections, we briefly describe the model simulation scenarios and the aircraft measurement programs, followed by results and discussions. Some conclusions are given in section 4.
2 Models and Measurements
2.1 Simulation Concentrations
 The institutes participating in the TransCom-CH4 experiment produced global CH4 distribution using 13 models and four variants with different spatial resolutions and OH distributions. The CTM simulation outputs include CH4 vertical profiles over various sites for a decadal period (1992–2007) with a 1 or 3 h time interval, for six CH4 tracers that used different surface emission scenarios. A detailed description of the TransCom-CH4 experiment is documented in Patra et al. [2011a]. In this study, we use CH4 profiles simulated by 12 CTMs (Table 1), namely, the ACTM (atmospheric general circulation model–based CTM) [Patra et al., 2009a, 2009b], CAM (Community Atmosphere Model) [Gent et al., 2009], CCAM (Conformal-Cubic Atmospheric Model) [Law et al., 2006], GEOS-Chem (Goddard Earth Observing System (GEOS) meteorology-driven CTM) [Fraser et al., 2011], IFS (Integrated Forecasting System) [www.ecmwf.int/research/ifsdocs/CY36r1/index.html], IMPACT_1×1.25 (Integrated Massively Parallel Atmospheric Chemical Transport) [Rotman et al., 2004], LMDZ (Laboratoire de Météorologie Dynamique—Zoom, version 4) [Hourdin et al., 2006], MOZART (Model for Ozone and Related Chemical Tracers. version 4) [Emmons et al., 2010], NIES08i (NIES CTM. version 2008 isentropic) [Belikov et al., 2011], PCTM (Parameterized CTM) [Kawa et al., 2004; Bian et al., 2006], TM5_1×1 (Transport Model, version 5) [Krol et al., 2005], and TOMCAT CTM [Chipperfield, 2006].
Table 1. List of the CTMs Used in This Study (See Text for Further Details) and Basic Model Features
Source of Meteorology
~2.8 × 2.8°
NCEP2; U, V, T; SST
2.5 × ~1.9°
NCEP; U, V; SST
2.5 × 2.0°
0.7 × 0.7°
1.25 × 1.0°
3.75 × 2.5°
ECMWF; U, V, T; SST
~1.8 × 1.8°
2.5 × 2.5°
1.25 × 1.0°
1.0 × 1.0°
~2.8 × 2.8°
 These selected models used reanalyzed meteorology for simulating synoptic and shorter time scale (<7 days) variations in the observed time series. The newly participating IFS model is based on a semi-Lagrangian semiexplicit version of the European Centre for Medium-Range Weather Forecasts (ECMWF) spectral model (i.e., Eulerian vorticity-divergence configuration). The IFS model is run at the 0.7° × 0.7° horizontal resolution, finest among all the participating models. The IFS model simulations of SF6 and CO2 has been found satisfactory in the TransCom continuous experiment [Law et al., 2008].
 Five models, namely, ACCESS (The Australian Community Climate and Earth-System Simulator) [Corbin and Law, 2011] driven by climatological meteorology, ACTM_OH (OH sensitivity run of ACTM), GEOS-Chem_DOH (OH sensitivity run), IMPACT (low-resolution version), and TM5 (low-resolution version), are excluded from main discussions of this analysis. For the control simulations, three-dimensional OH fields at monthly mean time intervals, based on Spivakovsky et al. , have been prescribed in the TransCom-CH4 protocol. Two of the models (ACTM and GEOS-Chem) repeated their TransCom runs with the same resolution and meteorology but with different OH fields. For both models, the vertical profile gradient was similar in the two different runs. The IMPACT and TM5 models repeated their TransCom runs using different horizontal resolution (implicitly different meteorology). For both models, we find relatively large differences, greater than the differences between different models, in some seasons. No clear distinction can be made whether the higher- or lower-resolution version of the models is better for these two models (see Figure S1 in the supporting information). However, earlier analysis has shown that the higher-resolution version of TM5 performs better than its coarser version for the interhemispheric gradients of SF6, CH4, and CH3CCl3 near the Earth's surface [Patra et al., 2011a].
 We analyzed CH4 tracers simulated using (1) the control emission scenario (CTL) (Figure 1) that combined seasonally varying natural sources (CYC NAT; major components being the emissions from biomass burning and wetlands) [see Fung et al., 1991] and interannually varying anthropogenic emissions (ANT_IAV; based on EDGAR3.2) [see Olivier and Berdowski, 2001; Patra et al., 2009a for details] and (2) the EXTRA flux, in which interannually varying biomass burning emissions are based on GFEDv3 [van der Werf et al., 2006] and wetland emissions are from the VISIT terrestrial ecosystem model [Ito and Inatomi, 2012]. Based on our experiences in Patra et al. [2011a], these two tracers are sufficient to evaluate the differences in the simulations caused by the emission distribution and strength. The CTL and EXTRA emissions differ by more than 100% over some of the regions (Figure 1b). To remove systematic offsets between the CTM simulations, the decadal mean concentrations in the altitude range of 2–6 km over Hawaii (HAA) is subtracted from the simulation results. This implies that the horizontal gradients between sites can be studied in reference to Hawaii, but the offset treatment does not affect the determination of vertical gradients or other parts of analysis. For all statistical calculations, the model values are sampled at the time and location of the observations. Model values, between the vertical layers below and above their respective coordinates, are linearly interpolated to measurement height.
2.2 Aircraft Measurements
 Vertical CH4 profiles have been measured using small aircraft by the National Oceanic and Atmospheric Administration (NOAA) Earth System Research Laboratory (ESRL) at sites in North America and the Pacific (www.esrl.noaa.gov/gmd/ccgg/aircraft/index.html), by the National Institute for Environmental Studies (NIES) at a site in Siberia [Machida et al., 2001], and by NOAA/ESRL in collaboration with Instituto de Pesquisas Energéticas e Nucleares at sites in Amazon [Gatti et al., 2010]. The site names are listed in Table 2, and their locations are plotted with green symbols on a global CH4 emission map in Figure 1a. Air samples are collected using aircraft in the altitudes from the Earth's surface up to 4–10 km depending on the site (Table 2). The aircraft measurements are conducted a few times a month with measurements continuing to the present for all sites, except for Amazon (SAN), where they terminated in 2005. The observation period for each site overlapping with our analysis period is shown in Table 2. Only 2 of the 13 sites are located in the Southern Hemisphere, and 8 of the 11 sites in Northern Hemisphere are located in and around the North American continent. Thus, our analysis of vertical and horizontal gradients does not cover the globe as uniformly as desired for evaluating vertical and horizontal gradients in CH4 and other trace gases.
Table 2. Site Details of the Aircraft Measurements Used in This Studya
Above Mean Sea Level (m)
The Surgut site is operated by NIES, and all other sites are by the NOAA/ESRL.
Southern Great Plains
 The samples collected over Surgut, Russia, are analyzed using gas chromatography (Agilent 5890, Agilent Technologies, Inc.) equipped with a flame ionization detector (GC-FID) calibrated against the NIES-94 CH4 scale. The analytical precision of the GC-FID for the CH4 concentration was estimated to be less than 2.0 ppb. The NIES-94 CH4 scale is higher than the WMO (NOAA-04) scale by 3.5–4.6 ppb in the concentration range of 1750–1840 ppb [Zhou et al., 2009]. To account for the calibration scale difference, an average value of 4 ppb is subtracted from the measurements over Surgut.
 The NOAA ESRL aircraft measurements are made by collecting air samples in 0.7 L silicate glass flasks at biweekly to monthly intervals at about 16 sites on a regular basis and on campaign mode over Santarem, Brazil (Table 2). Air samples are collected using a turboprop aircraft in the altitude range of about 500 m to 8 km. CH4 is measured by GC-FID at accuracy of ±1.2 ppb. The concentrations are reported at the absolute calibration scale of NOAA-04 that is maintained as the WMO standard [Dlugokencky et al., 2005].
3 Results and Discussions
 We compare the CTM simulation results with the aircraft measurements to understand the differences between the models and evaluate performance of the models in comparison with measurements during the analysis period of 2001–2007.
3.1 Seasonal and Daily Variations at Different Heights
 Figure 2 shows the observed and simulated time series of CH4 concentration for 2007 for three different height ranges (0.5–2, 3–5, and 6–8 km) and six sites. Information on seasonal cycles is sometimes difficult to extract from the measurements given the irregularity in data coverage and gaps in the time series (black dots). For better clarity, we only show multimodel mean seasonal cycles (black line), which are created by fitting a curve (three harmonics) to the afternoon (1200–1600 h local time) mean time series. This method ensures unambiguous identification of the peaks and troughs in the simulated seasonal cycles of CH4, which is otherwise not well defined due to large data gaps at some of the observation sites. The amplitude and phase of the simulated seasonal cycles are similar for most models, e.g., except TOMCAT, at the northernmost site (PFA). Although the synoptic and daily scale variations show differences between the modeling results, we find that the observed seasonal cycle matches the models within 1σ standard deviations at most sites (Table 3). The observed seasonal cycle is significantly different from the model mean seasonal cycles for some of the continental/coastal sites (ULB, HFM, TGC, and SAN) at 1–3 km height, where the uncertainties in both surface emissions and model transport are greater compared to those at the remote marine site. The synoptic variations (2–10 day) are consistent between models (correlation coefficients generally greater than 0.5; Table S1), indicating similarities in reanalyzed meteorological parameters used in the models. While the behavior of seasonal cycles depends on temporally varying regional emissions and seasonality in zonal mean OH radical, the synoptic variations in CH4 are produced by regional patterns of emissions and model transport variability at the synoptic time scale. Therefore, we conclude that the synoptic transport and CH4 emission distribution are fairly well represented in all models.
Table 3. Measured and Model Simulated Seasonal Cycle Amplitudes of CH4 at Two Altitude Rangesa
Model results are given as mean and 1σ standard deviations (individual model results are available in Table S1).
 The peak-to-trough seasonal cycle amplitude at 1–3 and 3–5 km height ranges for SAN, TGC, SGP, and SUR are relatively large compared with the other sites (Tables 3 and S1). Greater daily CH4 variations during summer are simulated at SUR, PFA, LEF, and ULB near the surface, coinciding with the seasonal maximum of CH4, and are thought to be caused by the increase in surface emissions in summer and the seasonally varying dynamics associated with the planetary boundary layer (BL) height as well as cumulus convection. On the other hand, troughs of the multimodel mean seasonal cycles at TGC, HAA, and the other remote sites sampling background air show lower variability during the boreal summer months, coinciding with the seasonal minimum of CH4 in the troposphere, except for RTA. In general, the seasonal amplitudes in the lower troposphere are similar to or larger than those in the upper troposphere. The seasonal amplitude at the northern central Pacific site HAA decreases from 36.9 to 30.2 ppb with altitude, whereas the seasonal amplitude at the southern central Pacific site RTA increases from 26.4 ppb at 1–3 km to 35.1 ppb at 3–5 km (Table 3). Throughout the Northern Hemisphere, the amplitude weakens upward as the emission signal from the lower troposphere propagates to the middle/upper troposphere. This weakening is especially remarkable at TGC (Figure 2e), where the CH4 seasonal cycle amplitude at the 6–8 km range is reduced to 11% of that at 0.5–2 km (all model average of 79 ppb). On average, the amplitude decreased only marginally at ULB PFA, ESP, and LEF, while at other sites, the amplitude decreased between 30% and 60% at 6–8 km compared to that at 0.5–2 km. Note here that the anthropogenic or biogenic emissions around the Mongolian site ULB is among the lowest observed over the land regions, while PFA, ESP, and LEF are located around the regions of strong CH4 emissions during the boreal summer.
 One unusual feature among the seasonal cycles is observed as an increase in CH4 at 6–8 km height over SUR (73°E, 61°N) during May–July. The timing of this peak in June–July at 6–8 km is ahead of the seasonal peak seen near the surface (0.5–2 km) in July–August, and there is almost no seasonal variation in the 3–5 km altitude. An examination of CH4 longitude-pressure cross sections (latitude range: 60°N–65°N; Figure S2) suggests that the high CH4 over SUR during the late spring/early summer is transported from Western Europe. Air masses carrying the signature of European emissions are lifted over the Ural mountain range (57°E, 51°N to 66°E, 68°N) and are subsequently transported to Siberia, Mongolia, and as far as western North America by the prevalent westerly winds, as suggested by the longitude-pressure cross sections of CH4 concentrations (Figure S2). The CH4 peak near the surface is coincident with the seasonal high in emissions from the wetlands in the Siberian region.
3.2 Vertical Profiles
 Figure 3 shows the seasonal mean vertical CH4 profiles at SUR, TGC, HAA, and SAN with the CTL fluxes during boreal winter (January–March: JFM) and summer (July–September: JAS). The modeled (color lines) and observed (black dots with horizontal bars) vertical profile variations with altitude agree well, inside the standard deviation (black bars) of the seasonal variation of the observed CH4 concentrations.
 The simulated vertical profiles at SAN show larger differences between models (a spread of 50 ppb at all altitudes) and with measurements. Due to sparse measurements at some of the sites, e.g., SAN, the significance of model-observation differences cannot be assessed here, and the altitude range is limited to below 4 km. CH4 concentrations near the surface are affected by biospheric or anthropogenic activities (including biomass burning) as well as meteorological conditions. For example, clear synoptic variability is observed near the surface at TGC during the JAS months as the site is either under the influence of marine air from the Atlantic Ocean from the southeast (seen as relatively low concentrations) or from regional emission hot spots around the Gulf of Mexico. During the JFM months, winds are predominantly southwesterly, and the TGC site is under the influence of emissions in Mexico and southwestern United States. Thus, the concentrations are relatively stable (lower variability), and the average values are higher by about 50 ppb in JFM months than those in the JAS months. This is due to slower chemical loss of CH4 by OH in JFM compared to the JAS months in this latitude band.
 At SUR, except below 2 km, the simulated results are generally lower than the observed CH4 concentrations during both seasons. A more contrasting feature is observed in the JAS season when the measurements show slightly increasing or fairly constant CH4 concentration between 2–3 and 7–8 km at SUR (and also at TGC). Such elevated levels of gases in the upper troposphere (7–8 km altitude), with emissions only on the Earth's surface, occur frequently in the presence of deep cumulus convection [e.g., Patra et al., 2009b] (refer also to Figure S2 for the cross-sectional view over Siberia). Only ACTM and CCAM simulations capture this feature to some extent over SUR. This is a clear indication that 10 out of 12 models underestimate vertical transport of surface emissions to the upper troposphere over Surgut due to deep cumulus convection during the JAS season. As discussed earlier, the higher CH4 values during May–June were under the influence of emissions over the Western Europe.
 The concentration at SAN sharply decreases with altitude in the lower troposphere, suggesting significant CH4 emission at this location, and the profiles above 3 km are relatively straight or slightly increasing in both seasons, indicating strong vertical motion in all seasons over this tropical site (Figures 3g and 3h).
 When the convective activity is intense during the boreal summer, relatively straight vertical profiles are also observed at HAA, TGC, and other NH sites. The profiles at ocean sites HAA and RTA usually reveal characteristics of a background site with insignificant surface emissions around the site during all seasons. However, emission signals from Equatorial Asia/Western Pacific and Temperate Asia reach HAA during October–January and April–July, respectively, at 850 hPa altitude (Figure S2). In addition, the Hawaiian Islands are located at the subtropical mixing barrier for chemical species [Miyazaki et al., 2008], causing greater seasonal variability at HAA than RTA in the lower troposphere (Table 3). Thus, the vertical profiles at HAA show a gradient in the JFM months despite being an ocean site due to the influence of northern hemispheric emissions and weak OH loss (Figure 3e).
 The CH4 peak at 3–4 km during the JAS months over HAA (Figure 3f), typical of all years, is caused by the transport of high CH4 from Boreal Asia along the isentropic surfaces when the convective activity is low. Four models (GEOS-Chem, PCTM, IFS, and TM5) are able to simulate this feature to some extent. GEOS-Chem and PCTM are driven by the meteorological reanalysis from the Goddard Earth Observing System Data Assimilation System, whereas IFS and TM5 use ERA-Interim meteorology from the European Centre for Medium-Range Weather Forecasts (ECMWF). It is however unclear whether the better simulations by four models using GEOS and ECMWF reanalysis are due to the quality of meteorological products or from the representation of transport in the models. Sensitivity simulations using one of the models with both reanalysis products are needed to address this issue.
3.3 Differences Between the Observations and Model Simulations
 Figures 4a–4d show the model-observation scatterplots at four different height ranges (black: 0.5–1.5 km, blue: 1.5–3.0 km, green: 3.0–5.0 km, and red: 5.0–7.0 km) at sites SUR, THD, SGP, and HAA. The black diagonal line intersects at the coordinate origin with slope 1. Correlation coefficients R between the model and observation time series are presented in Table S1. Several values, denoted by “0,” did not satisfy a significance level of 0.05 based on the critical value for testing the null hypothesis H0 of noncorrelation between two time series. Higher correlation coefficients in 3–5 km altitude range than that at 1–3 km altitude range suggest that the variance of model-observation difference is larger near the surface than in the free troposphere at most sites. This is primarily due to large emission signals as well as greater site representation error near the Earth's surface at synoptic time scales. These emissions are uncertain (e.g., monthly averages are used), which leads to larger model error near the surface. In contrast, the signals in the midtroposphere are dominated by transport processes.
 Figures 4e–4h shows the frequency distribution of the model-observation difference at different seasons in winter (black: JFM), spring (blue: AMJ), summer (green: JAS), and autumn (red: OND) for the analysis years of 2001–2007. The model-observation difference at HAA in Figure 4h (also CAR, RTA, and ULB) falls within a small range of CH4 concentrations (±15 ppb, 1σ), and has a higher correlation (generally greater than 0.7 in Table S1, except NIES08i and TOMCAT) than at the other sites because of their location in the background marine environment (RTA and HAA) and inland (CAR and ULB) away from direct influence of continental sources (Figure 1). The model-observation differences at SUR, SGP, and THD (also SAN, LEF, HFM, and PFA) are as large as 100 ppb for the high CH4 concentrations near the Earth's surface (Figure 4).
 The frequency distribution for the small model-observation mismatches in the regions around HAA (Figure 4h), CAR, RTA, and ULB (not shown) shows little or no skewness. The low (high) skewness indicates small (large) seasonally varying model biases, while high kurtosis suggests that the model time series is well correlated with the observation time series. The skewness at THD is also small, and the kurtosis is low compared with the Gaussian distribution that is symmetric around 0 difference. On the other hand, SGP shows negative skewness, and the peaks of the population densities in spring and summer (blue and green histograms in Figure 4g) are negatively biased by about 20 ppb. Thus, Figures 4c and 4g confirm that the models simulated relatively low CH4 concentrations compared to the observations at SGP. At SUR, the peak in summer (green in Figure 4e) is marginally skewed positive due to an overestimation of the modeled CH4 concentrations. These results suggest that the used CTL CH4 emissions are too low (marginally high) around the SGP (SUR) site.
 These biases in simulated CH4 concentrations are unlikely to be due to transport model errors, because most models systematically show similar underestimations/overestimations, and the performance of the TransCom-CH4 model ensemble has been found to be satisfactory at all other sites (e.g., Figures 4f, 4h, and S3). In addition, the lifetime of atmospheric CH4 is longer than 1 year in the troposphere, much longer compared to the transport or emission time scales around any given location [Patra et al., 2009a]. Thus, the model biases between sites within a hemisphere/continent are not likely to arise from uncertainty in OH abundance. Figures 5a and 5b show mean model-observation differences due to two emission scenarios used in this analysis, CTL (blue symbols) and EXTRA (red symbols), in the middle and lower tropospheric altitudes. Smaller model-observation differences are found at SUR in the EXTRA simulations, which has lower CH4 emissions compared to CTL case in the Siberian region (Figure 1b). This suggests that the CTL CH4 emissions around SUR are too high. Model simulations using either CTL or EXTRA emissions underestimate the observed CH4 concentrations at SAN, suggesting stronger emissions from the Amazon region or the vertical transport is faster in the models.
3.4 Vertical, Latitudinal, and Longitudinal Gradients
 We show the annual mean simulated and observed vertical gradients in Figure 5c and Table S2. Here the vertical gradient in CH4 profiles is defined as the difference of concentrations between the 1–3 km and 4–6 km heights. The simulated vertical gradients are generally within 5 ppb of the observed gradients at the oceanic sites (i.e., RTA and HAA), most of the coastal sites (i.e., TGC, ESP and PFA), and some of the continental sites (i.e., CAR, SUR, and ULB). The differences between simulated and observed vertical gradients are more than 5 ppb at SAN, SGP, LEF, and HFM (continental sites). The model-observation differences are systematically biased positive and negative, respectively, at SUR/THD and SAN/SGP, for most of the transport models. These results are consistent with overestimation and underestimation of simulated concentrations compared to observations in the lower troposphere at SUR and SGP, respectively (Figures 4a and 4e and Figures 4c and 4g). These systematic model-observation biases are thus likely to be caused by regional emissions in the CTL scenario rather than errors in CH4 loss (longer than 1 year lifetime due to OH reaction) and model transport. By using EXTRA emissions (red symbols), the simulated vertical gradients worsen for the sites SAN, HFM, and ULB, in comparison with the observations, but improve slightly over SUR.
 Figure 6 shows comparisons of the simulated and observed latitudinal distributions of CH4 concentrations among all sites at different height ranges (1.0–2.5, 2.5–5.0, and 5.0–8.0 km). The measured CH4 concentrations increase from SH to NH between 20°S and 50°N in the troposphere, at near linear rates of 1.4 ppb/deg at 1.0–2.5 km, 1.3 ppb/deg at 2.5–5.0 km, and 1.2 ppb/deg at 5.0–8.0 km. These values are generally well simulated by the models using both CTL and EXTRA emission scenarios. The differences in latitudinal gradient are mainly due to the decrease in CH4 concentrations with height in the NH, e.g., the upper tropospheric concentration in the NH high-latitude sites is ~30 ppb lower than the lower tropospheric values. The agreement between model and observations is best for the 2.5–5.0 km altitude range, where the standard deviation of model-observation differences is ~5 ppb (see also Table S3). Although model uncertainty in the free troposphere is usually small, variance among the models at ULB at 5 Alkm is larger than that in the other sites. This is presumably due to the long-range transport of European and Siberian CH4 emission signals to ULB as discussed earlier (section 3.1). The smaller (1.2 ppb/deg) interhemispheric gradient at 5intkm, compared to that at 2.5–5.0 km heights, indicates that the CH4 in the NH upper troposphere is more readily transported to the SH. This effect can be seen as the higher CH4 concentration in the upper troposphere at RTA than in the lower troposphere. A higher concentration in the SH upper troposphere compared to that in the lower troposphere is also observed for nonreactive species, such as CO2 and SF6 [Nakazawa et al., 1991; Miyazaki et al., 2008; Patra et al., 2009b; Sawa et al., 2012]. The role of surface emissions on meridional gradient weakens rapidly with increasing height between 1.0–2.5 and 2.5–5.0 km (Figures 6b and 6c).
 Figure 7 shows the variation of CH4 with longitude for sites north of 42°N (i.e., SUR, ULB, PFA, ESP, LEF, and HFM) in the same height ranges as Figure 6 in order to investigate zonal CH4 transport and vertical propagation driven by surface CH4 emissions from North America, Europe, and Siberia. The lower tropospheric variations in CH4 are mainly linked to stronger emissions around SUR and HFM compared to those around ULB and ESP. The CH4 concentrations in the middle troposphere clearly have an eastward decreasing trend in observation and models for both CTL and EXTRA emissions (about −0.1 ppb/deg from the prime meridian to eastward; Figure 7b). The high CH4 concentration in the lower troposphere over Siberia (Figure 7c) is diluted through mixing and reduces the difference between average values at SUR and other sites (except HFM) at the 5–8 km height range (Figure 7a). This behavior is seen consistently for measurements and model using both the emission scenarios. Apparently, the CH4-rich air over the Eurasian continent in the upper troposphere extends to the western side of the North American continent (PFA and ESP) as suggested by the longitude-pressure cross-sectional plots (refer to Figure S2). The air in the upper troposphere on the eastern side of northern United States mostly originates at about 40°N over the Pacific Ocean, except for the winter months, giving relatively low CH4 concentrations at HFM. These results are generally consistent with Xiong et al. , but our understanding is improved significantly on the longitudinal variations in upper tropospheric CH4 by including a greater number of vertical profile measurements.
3.5 Correlation of Latitudinal and Vertical Transports
 The overestimations or underestimations of latitudinal gradients of long-lived atmospheric species simulated by models, compared to the measurements, are thought to be linked with the vertical concentration gradients of those species [Denning et al., 1999]. To explore this hypothesis further, we show scatterplots of vertical gradients at each site and latitudinal gradient, with reference to HAA, within two altitude ranges 3–5 and 0.5–2.5 km (Figure 8). Both gradients are plotted as the differences relative to the measured gradients, showing which models overestimate or underestimate the vertical and latitudinal gradients. Relative gradients are chosen to remove biases in simulated concentrations with respect to measurement, which are often greater than the measured gradients. In the middle troposphere (3–5 km), relative vertical and latitudinal gradients are scattered in the approximate ranges of −10 to +7 ppb/km and −1.5 to +2.0 ppb/deg, respectively. Loose correlation between vertical and horizontal gradients suggests that the interhemispheric transport is relatively insensitive to the local/regional vertical profile of CH4 in the 3–5 km altitude range and that the transport in the models is similar. However, the relative latitudinal gradients are apparently proportional to the relative vertical gradient in the lower troposphere (0.5–2.5 km), suggesting that the vertical transport of emission signals through the lower troposphere plays a critical role in determining interhemispheric gradients near the Earth's surface.
 A schematic representation of the relationship between vertical and interhemispheric transports is shown in Figure 9a. The monthly mean modeled vertical gradients in the NH are calculated as the difference between the CH4 concentrations between 900 and 500 hPa altitudes. Values are also averaged over all the model grid cells between 30°N and 50°N for 2007. Average latitudinal gradients are calculated as the difference between mean CH4 for SH (50°S−30°S) and NH (30°N −50°N). Annual mean values of the interhemispheric and NH vertical gradients in CH4 show a compact relationship (black symbols). The seasonal variations in this relationship are mainly controlled by the vertical transports and the emission and loss rates of CH4. In July, both the latitudinal and NH vertical gradients are smaller than those in the other seasons because of faster loss rates in the NH and lower loss rates in the SH as well as stronger vertical mixing in the NH, despite high emission rates in the NH. In October, the NH vertical gradients are the largest among all seasons (high emission rate, low loss scenario), although the latitudinal gradients do not increase significantly. In January, the midtropospheric CH4 concentrations reach a seasonal high because of accumulation in the NH through autumn and early winter (low emission, very low loss scenario, and weaker vertical mixing). This condition leads to the largest interhemispheric gradients in January.
 The interactions between vertical and latitudinal gradients in CH4 (also valid for tracers in the troposphere with lifetimes greater than the interhemispheric transport time of about 1.3 years; see Figure S5 for SF6 results) are schematically depicted in Figure 9b. Based on our analysis, we hypothesize that (1) vertical transport is driven by the strength of convection as well as turbulent mixing in the boundary layer (BL) and (2) meridional gradients depend mainly on the mixing in the BL for sites close to emissions and on the horizontal advection scheme for the remote/background sites. However, there is interaction between the turbulent mixing, convection, and advection processes in the model. A study using ACTM [Patra et al., 2009b] suggested that the turbulent mixing transports emission signals to the 900–800 mb height layer, and the convective processes transport species from the BL to the middle and upper troposphere (MUT). Most of the interhemispheric transport due to advection occurs at higher altitudes. Thus, a model transport with weaker turbulent mixing and convection has a tendency toward steeper meridional gradient in the simulated concentrations at lower altitudes but not at higher altitudes (6–8 km).
 As discussed in Patra et al. [2011a] and shown in Figure 9a, the TM5 model produces one of the largest latitudinal gradients in CH4 among all the models that participated in the TransCom-CH4 experiment. However, we find in this study that the NH vertical gradients simulated by TM5 are not the largest. The latitudinal gradients shown in Figure 6 suggest that the stronger gradients in the TM5 simulations are mainly due to low SH concentrations compared to other models, likely to be caused by slower meridional transport. PCTM produced the largest vertical gradients below 1–2 km (refer to Figures 3 and 9a), most likely due to trapping of emissions within the planetary boundary layer. However, the simulated interhemispheric gradients are not extreme because stronger vertical gradients for PCTM in both the hemispheres (Figure 5c) result in an interhemispheric gradient that is close to the multimodel average. All other models suggest that a stronger vertical gradient in NH troposphere corresponds to a stronger interhemispheric gradient in CH4 concentrations.
 We assessed zonal, meridional, and vertical transports of tropospheric CH4 using the chemical transport models (CTMs) and two emission scenarios (CTL and EXTRA) in comparison with aircraft measurements. The model-model difference is large near the surface in source regions and during strong convection and small in the middle/upper troposphere and over the marine background regions. The model-observation differences are controlled predominantly by the model transport and the surface emission rather than the loss rates of CH4. At most sites, the simulated concentrations vary around the observed values, while at three sites, the model results are systematically biased relative to the observed value. The positive (at Surgut, Russia) and negative (at South Great Plain, United States, and Santarem, Brazil) biases, systematic for most models, can be attributed to higher and lower CH4 emissions as estimated in the CTL case. These model-observation comparison results are generally consistent for simulations using both the emission scenarios, CTL or EXTRA (except for Surgut, where simulations using lower emissions over the Siberian region, as in EXTRA, showed better model-observation agreements).
 Observed CH4 profiles show near-zero vertical gradients above 3 km, where models indicate that the convective activity is the primary driver of mixing (all seasons over the Amazon and summer months in the higher latitudes), and elevated concentrations in the upper troposphere region, when measurements are available over Surgut. In contrast, stronger decreases with altitudes above 3 km are observed in the middle- and high-latitude land regions, where the anthropogenic emissions are intense (e.g., TGC, HFM, and SUR) compared to the lower emissions over the background marine sites (e.g., HAA and ULB). The model-model differences are found to be greatest at sites located over intense emission regions for the height below 3 km. In the middle and upper troposphere (MUT), greater model-model differences are found over the regions with stronger convective activities (e.g., SAN and NH sites during summer). These two results suggest that the turbulent mixing (acting below 3 km) and deep cumulus convection (acting in the MUT region) parameterizations in the CTMs are less similar compared to the other transport processes, such as advection (except for one model).
 Interhemispheric gradients of the simulations and observations were 1.5 ± 0.1 and 1.3 ± 0.2 ppb/deg for the lower and middle troposphere, respectively. In an effort to understand the origin of model-model differences in interhemispheric gradients, we have shown scatterplots of vertical gradients in CH4 concentrations between lower and middle troposphere and latitudinal gradients between the two hemispheres. Generally, the models that produced higher vertical gradients for CH4 (and also SF6) concentrations also produced steeper interhemispheric gradients near the surface. In models with underestimated vertical mixing mainly due to boundary-layer turbulent mixing and parameterized convection, tracers tend to get trapped near the sources in the lower troposphere. In this case, the model is likely to overestimate meridional gradients in tracer concentrations, resulting from relative inefficient advective transport between the hemispheres (except for two exceptions in PCTM and TM5).
 This work was supported by JSPS/MEXT KAKENHI-A (grant 22241008). A. Fraser was supported by the UK Natural Environment Research Council National Centre for Earth Observation. We acknowledge the work of J. McGregor and M. Thatcher in the development of CCAM. CCAM simulations were undertaken as part of the Australian Climate Change Science Program and used the NCI National Facility in Canberra, ACT, Australia. R. Prinn and M. Rigby were supported by NASA-AGAGE grants NNX07AE89G and NNX11AF17G to MIT. M. Rigby was also supported by a NERC Advanced Fellowship. The TOMCAT work at the University of Leeds was supported by NERC/NCEO. The research leading to the IFS results has received funding from the European Community's Seventh Framework Programme [FP7 THEME (SPA.2011.1.5-02)] under grant 283576 in the context of the MACC-II project (Monitoring Atmospheric Composition and Climate—Interim Implementation). We sincerely thank all three anonymous reviewers for critical evaluation and providing very helpful comments and suggestions for improving the article.