Boundary layer sources for the Asian anticyclone: Regional contributions to a vertical conduit



[1] The transport of air from the planetary boundary layer (PBL) into the Asian Summer Monsoon anticyclone is investigated using backward trajectories initiated within the anti-cyclone at 100 mb and 200 mb during August 2011. Transport occurs through a well-defined conduit centered over the southern Tibetan plateau, where convection lofts air parcels into the anticyclone. The conduit, as a dynamical feature, is distinct from the anticyclone. Thus, while the anticyclone influences transport through the upper troposphere and lower stratosphere, it does not by itself define a transport pipeline through that region. To quantify model sensitivities, parcel trajectories are calculated using wind fields from multiple analysis data sets (European Centre for Medium-Range Weather Forecasts, National Center for Environmental Prediction's Global Forecasting System, and NASA's Modern-Era Retrospective Analysis for Research and Applications [MERRA]) and from synthetically modified data sets that explore the roles of vertical motion and horizontal resolution for discrepancies among these calculations. All calculations agree on the relative contributions to PBL sources for the anticyclone from large-scale regions with Tibetan Plateau and India/SE Asia being the most important. However, they disagree on the total fraction of air within the anticyclone that was recently in the PBL. At 200 mbar, calculations using MERRA are clear outliers due to problematic vertical motion in those data. Large differences among the different data sets at 100 mbar are more closely related to horizontal resolution. It is speculated that this reflects the importance of deep, small-scale convective updrafts for transport to 100 mbar.

1 Introduction

[2] The overriding goal of this study is to better understand the dynamical processes that transport air from the planetary boundary layer (PBL) into the Asian summer monsoon (ASM) anticyclone (AAC) and ultimately into the stratosphere. The AAC lies at the northern boundary of the tropical tropopause layer (TTL), a transition layer that merges the upper tropical troposphere with the lower stratosphere. The TTL, as defined by Fueglistaler et al., [2009], is situated in the pressure range 70–150 mbar (altitude range ~14–19 km) in latitudes bounded by the subtropical jets and is the primary gateway through which tropospheric air enters the stratosphere. As such, dynamical circulations within the TTL exert a strong influence on the transport of important chemical constituents from their source regions in the PBL into the stratosphere. Vertical motion in the troposphere below the TTL is characterized by large-scale downwelling interrupted by strong, but relatively small-scale, convective updrafts. The TTL itself is generally characterized by weak large-scale upwelling with occasional convective intrusions. However, the lower altitudes of the TTL also contain regions of large-scale downwelling [e.g., Bergman et al., 2012]. Thus, in order for boundary layer air to be transported efficiently into the stratosphere, it must survive two important filtering mechanisms. First, it must be transported into the TTL via deep convection, and then it must be carried by favorable horizontal winds through regions of upwelling.

[3] Anticyclones have three characteristics that make them potentially important for vertical transport through the TTL: (1) They are associated with anomalously strong meridional winds that promote exchange of air between the troposphere and midlatitude stratosphere [Dunkerton, 1995; Chen, 1995, Dethof et al., 1999]. (2) They are the classical linear response to deep convection [Gill, 1980; Matsuno, 1966; Hoskins and Rodwell, 1995] and thus are always in close proximity to deep tropical convection, which is the vehicle that carries boundary layer air into the TTL. (3) The closed streamlines associated with anticyclones promote strong potential vorticity gradients that isolate air within the anticyclone from that outside, keeping air confined to the anticyclone for prolonged periods [Li et al., 2005; Randel and Park, 2006]. In the case of the ASM anticyclone, this confinement keeps air parcels away from the coldest temperatures in the TTL, reducing the dehydrating influence of these cold traps [e.g., Bannister et al., 2004], and it prolongs their exposure to elevated heating rates associated with clouds over the Asian monsoon sector [Bergman et al., 2012], which promotes efficient vertical transport through the TTL [e.g., Corti et al., 2006].

[4] The ASM anticyclone is in a remote region of the atmosphere where few in situ measurements are available. Nevertheless, it has a well-established influence on the transport of air from the tropical troposphere into the lower stratosphere. Satellite measurements show that air in the Asian anticyclone has anomalously high concentrations of trace constituents associated with tropospheric air, such as methane and nitrogen oxides [Park et al., 2004], CO [e.g., Park et al., 2007], and HCN [Randel et al., 2010], and anomalously low concentrations of stratospheric constituents such as ozone [Randel et al., 2001; Gettelman et al., 2004; Park et al., 2007]. The ASM anticyclone also has anomalously high concentrations of water vapor, suggesting that it might be important for transporting moist air from the ASM [e.g., Gettelman et al., 2004; Fu et al., 2006; Park et al., 2007, Bian et al., 2012] into the stratosphere. While there are continuing efforts to understand how the Asian monsoon sector contributes to the chemical constituents of air entering the stratosphere [e.g., Dethof et al., 1999; Bannister et al., 2004; Park et al., 2009; Chen et al., 2012], the dynamics are complex and there are many open questions including the focus of this study—what are the primary source regions for air within the ASM anticyclone?

[5] Lagrangian particle models that calculate air parcel trajectories are well suited for investigating transport dynamics and, as such, are commonly used for investigating transport through the TTL and water vapor distributions in the stratosphere [e.g., Jackson et al., 2001; Legras et al., 2003; Bannister et al., 2004; Bonazzola and Haynes, 2004; Jensen and Pfister, 2004; Fueglistaler et al., 2004; James et al., 2008; Liu et al., 2010; Ploeger et al., 2011; Tzella and Legras, 2011; Bergman et al., 2012]. While precise tracking of air parcels is not feasible for times typical of transport through the TTL (30–60 d), Lagrangian calculations can, in principle, supply probability distributions from which useful statistical measures of air transport can be derived. In particular, an ensemble of “back” trajectories, which trace parcel paths backward in time from a target location x0 and time t0 to a source (xsrc,tsrc), can yield a statistical description of the conditions that determine the atmospheric state at (x0,t0).

[6] Lagrangian particle models require detailed “forcing” fields, which are typically obtained from analyzed meteorological data sets, to determine the air parcel trajectories. There are two common approaches: “kinematic” trajectories are derived in pressure coordinates and use pressure velocity ω to calculate vertical displacements, while “diabatic” trajectories are derived in isentropic coordinates using diabatic heating rates math formula to calculate vertical displacements. The best choice among these approaches depends on the application. For nearly adiabatic flows, such as in the upper troposphere and stratosphere, the diabatic approach is often preferred because vertical motion is determined by small deviations to a largely horizontal circulation. In these cases, the diabatic approach is less prone to error amplification than the kinematic approach [e.g., Danielsen, 1961; Schoeberl et al., 2003; Ploeger et al., 2010, 2011; Schoeberl and Dessler, 2011]. For tropospheric flows, particularly in the tropics, small vertical gradients of potential temperature and large diabatic heating rates reduce this advantage.

[7] In remote regions such as the tropics, the scarcity of in situ measurements introduces large uncertainties to the analyzed meteorological data. This is particularly true for the vertical motion fields ω and math formula, which are not directly observed and are, instead, model-derived quantities. In light of these considerations, it is interesting that the importance of the ASM for the transport of air into the tropical stratosphere as determined by trajectory analysis seems to depend on trajectory approach used; studies using the diabatic approach [James et al., 2008; Tzella and Legras, 2011; Wright et al., 2011; Bergman et al., 2012] emphasize the importance of the Asian monsoon sector whereas the kinematic approach apparently de-emphasizes it somewhat by promoting the western Pacific and Intertropical Convergence Zone (ITCZ) as additional source regions [Bonazzola and Haynes, 2004; Fueglistaler et al., 2004; Fueglistaler et al., 2005].

[8] In addition to potentially misleading results due to inaccuracies in the resolved dynamical fields, dynamical circulations that are not resolved by analyzed data sets can also impact the accuracy of Lagrangian trajectories, particularly in the ASM sector where small scale (i.e., a few kilometers or less) convective updrafts can transport air from the boundary layer to the upper troposphere within hours. Ideally, the resolved vertical motion represents the average over unresolved winds, which means that explicit representation of uncharacteristically slow updrafts and downdrafts are missing from the resolved fields in addition to fast updrafts. For some applications, the unresolved motions are important. For example, transport on the shortest time scales will be underrepresented as will the transport associated with isolated events such as tropical cyclones and volcanoes. However, the ubiquity of convective activity during the ASM can allow transport over longer time scales to be well represented by the resolved fields. While resolved fields will not allow parcels to be transported to the upper troposphere within hours by a single small-scale updraft, those parcels can nevertheless make their way to the upper troposphere over several days via the aggregate effect of ensembles of convective updrafts. While it is not possible to definitively distinguish between these two situations without fully resolved convection, we can probe the robustness of our results through experiments that test their sensitivity to changes of model resolution.

[9] To better understand the dynamical impact of the ASM anticyclone on vertical transport through the troposphere, this paper analyzes the PBL sources and pathways for air entering the AAC. To do so, we calculate kinematic back trajectories that have been initialized within the anticyclone at 200 mbar and 100 mbar and follow them to their first encounter with the PBL. In section 3, we use this approach to map geographical locations for these sources as well as define the meteorological conditions under which they are transported into the anticyclone. Since our trajectories are initialized only within the AAC, the source regions we map are not necessarily the most important PBL source regions for the summer stratosphere; those regions have been identified, at least in part, in previous studies [e.g., Fueglistaler et al., 2005; James et al., 2008; Tzella and Legras, 2011; Bergman et al., 2012; Chen et al., 2012]. In section 4, we quantify the relative contributions to air within the anticyclone made by specified regions. This serves two purposes: it allows us to synthesize our results into manageable components, and it provides the statistical foundation for sensitivity experiments. Such experiments allow us to evaluate the robustness of our results and, in one case, actually allow us to determine which trajectories are the most realistic.

2 Experimental Design, Models, and Input Data

2.1 Experimental Design

[10] We perform kinematic back trajectory calculations with a Lagrangian particle model to analyze the boundary layer sources for the ASM anticyclone. Trajectories are initialized within the AAC at 200 mbar and at 100 mbar and tracked until they encounter the PBL or for 30 d (60 d for some calculations), whichever occurs first. To expedite the use of multiple forcing fields for these calculations, our analysis is confined to conditions observed during August 2011. This constrains our ability to draw general conclusions from our results, and further calculations that probe interannual and intraseasonal variability will be important. Nevertheless, since the ASM and anticyclone are robust features of the general circulation that occur regularly each year and are strongly tied to orography that fixes their geographical locations, the large-scale features identified here are expected to be robust as well. For all analyses presented here, the top of the PBL is defined according to local pressure

display math(1)

where, pPBL is the pressure at the top of the boundary layer, ps is surface pressure, λ is longitude, ϕ is latitude, and t is time. This definition places the top of the PBL at approximately 1.5 km above the surface. Given the breadth of knowledge regarding boundary layer dynamics, this definition of the PBL height is simplistic. However, it provides us with a value that places an air parcel close to the physical process that mix surface emissions into the atmosphere while avoiding the complications of modeling the impact of turbulence within the PBL on air parcel trajectories. This definition introduces a certain amount of uncertainty in identifying precise surface emission sources for the ASM anticyclone, but it also reduces the uncertainty introduced by problematic boundary layer parameterizations used to produce analyzed meteorological fields.

[11] Parcel trajectories are initialized every day during August 2011 (four times daily for some calculations) spaced every 1° in latitude and longitude in the region 15°N–45°N, 25°E–135°E. To select only parcels within the anticyclone, we apply geopotential height thresholds ΦAAC (12.52 km for parcels initiated at 200 mbar; 16.77 km for parcels initiated at 100 mbar) to each trajectory at its initial position and time. Trajectories with initial geopotential heights less than the thresholds do not contribute to the statistics shown in this paper. These thresholds were selected based on 4 years of 6 hourly data during August from the National Center for Environmental Prediction's Global Forecasting System (NCEP-GFS) operational analysis using criteria illustrated in Figure 1. This figure displays the geopotential height Φ within 40° of the regional maximum along a line of constant latitude at 100 mbar (Figure 1a) and at 200 mbar (Figure 1b). Each of the colored lines represents a single zonal profile; the thick black line is the 4-year average math formula. The horizontal line in each panel represents the threshold

display math(2)

where math formula is the maximum and math formula is the minimum value of math formula. With this choice of threshold, the AAC is essentially always present during August and the average width (zonally) is approximately 60°.

Figure 1.

Supporting data for the choice of geopotential height threshold used to identify the ASM anticyclone. Shown are zonal profiles of geopotential height through the anticyclone from the 6 hourly GFS operational analysis (thin color lines) during August 2007 to 2010 at (a) 100 mbar and (b) 200mbar. Center values (longitude 0) represent the maximum geopotential height within the AAC region (0°–60°N; 50°E–120°E). The remaining values for each profile represent geopotential height within ±40° longitude of the maximum along a line of constant latitude. The time-averaged profiles are indicated by thick black lines; thin black lines mark the heights of the geopotential thresholds.

[12] The average geographical location of the anticyclone during August 2011 is shown in Figure 2 in terms of the fraction of 6 hourly data points for which the geopotential threshold ΦAAC is exceeded. Close scrutiny reveals the slight northward shift of the anticyclone at 100 mbar (Figure 2b) compared to 200 mbar (Figure 2a) that is typical for the ASM anticyclone [Park et al., 2007; Bian et al., 2012]. The areal coverage of large values (e.g., dark shades) indicates the degree to which the position of the anticyclone is stationary. For example, the region near 30°N in the longitude range 50°E–100°E was within the 200 mbar anticyclone over 80% of the time during August 2011 (black shading in Figure 2a), whereas excursions of the anticyclone to encompass eastern China or northwest Africa were much less frequent.

Figure 2.

The August 2011 mean location of the ASM anticyclone at (a) 200 mbar and at (b) 100 mbar. Contoured are the fractions of 6 hourly geopotential height values that exceed the thresholds shown in Figure 1. Contour levels were chosen so that the locations of 90% of the trajectories initiated within the anticyclone are within the shaded regions and 20% are within the black shaded regions.

2.2 The Trajectory Model

[13] We use the Lagrangian particle model of Bergman et al. [2012], which is a slightly modified version of the Schoeberl and Sparling [1995] model. The model uses linearly interpolated winds from analyzed meteorological fields to advance the parcel trajectories back in time with the fourth-order Runge-Kutta method at a time step of 22.5 min. With the exception of two ensembles discussed at the end of section 4, all trajectories are calculated with the kinematic approach, using ω to determine vertical motion.

2.3 Input Data

[14] Forcing fields for the trajectories are obtained from three sources: (1) the operational model analysis from European Centre for Medium-Range Weather Forecasts (ECMWF); 6 hourly data at a horizontal resolution of 0.125° on 25 pressure levels (15 levels from 1000–100 mbar and 10 levels at pressures < 100 mbar); and published by the CISL Data Support Section at the National Center for Atmospheric Research, Boulder, Colorado, available online at; (2) the GFS operational global analysis; 6 hourly data at a horizontal resolution of 1° on 26 pressure levels (21 levels from 1000–100 mbar and 5 levels at pressures < 100 mbar); published by the CISL Data Support Section at the National Center for Atmospheric Research, Boulder, Colorado, available online at; and (3) the Modern-Era Retrospective Analysis for Research and Applications (MERRA) reanalysis data set [Rienecker and co-authors, 2008; Rienecker and co-authors, 2011]; 3 hourly data at 1.25° horizontal resolution on 42 pressure levels (25 levels from 1000–100 mbar and 17 levels at pressures < 100 mbar); data available from Goddard Earth Sciences Data and Information Center. All three data sets are derived with data assimilation systems that incorporate atmospheric general circulation models. The important distinguishing features are that the operational analyses are designed to provide initial conditions for weather forecasting and reanalysis data are designed for climate studies. Thus, error statistics for operational analyses are expected to improve with time, whereas reanalysis data are expected to have stationary error statistics.

3 Model Results: The Conduit to the Anticyclone

[15] In this section, we analyze the transport pathways taken by parcels from the PBL to the ASM anticyclone and relate those paths to the local meteorological conditions. Here pathways are revealed via horizontal cross sections of trajectory densities for those trajectories with PBL sources. Since it is possible for a trajectory to cross a specified horizontal surface multiple times, we quantify trajectory densities in terms of their “first” encounter (in reverse time) with the surface, just as boundary layer sources are defined in terms of the first encounter of a trajectory with the PBL. This provides a meaningful description of parcel pathways in two ways. First, it designates a unique position at each pressure surface for each trajectory with a PBL source. Second, it identifies the region for which parcels are unquestionably traveling upward on their way to the anticyclone and, so, identifies a region in which air parcels must be located in order to reach the anticyclone. In contrast, the collection of all surface crossings can contain regions where uplift to the anticyclone is aborted and even downward surface crossings and, so, only identifies the region in which air parcels that reach the anticyclone could have been located.

3.1 Experimental Design

[16] Ensembles of trajectories for this analysis are initialized four times daily (0Z, 6Z, 12Z, 18Z) throughout August 2011. Thirty day trajectories are initialized at 200 mbar and 60 d trajectories at 100 mbar. The extra integration time for the 100 mbar trajectories compensates for the additional vertical distance to the PBL and makes the fraction of parcels that encounter the PBL (81.4%) comparable to that for the 30 d 200 mbar trajectories (78.4%). Forcing fields for these trajectories are obtained from the ECMWF operational analysis data; however, due to the expense of using the full resolution data, the horizontal resolution of the wind fields is reduced to 1° with a cosine-smoothing window. Comparisons with a subset of trajectories calculated with the full resolution data (30 d trajectories initialized daily) indicate that the change of resolution affects primarily the small-scale structures that are difficult to interpret and are deemphasized here by 5° spatial smoothing of the fields displayed in figures. The impact of the resolution change on large-scale structures is subtle and does not alter any of the conclusions drawn in this section. As discussed in section 2, trajectories are initially spaced 1° longitude by 1° latitude in the region 25°E–135° E and 15°N–45°N and then filtered to exclude trajectories that do not meet the geopotential threshold. This results in a total statistical base of 114,011 trajectories for the analysis of the 200 mbar anticyclone and 99,891 trajectories for the 100 mbar analysis.

3.2 The Vertical Conduit

[17] To understand the prevailing meteorological conditions that guide parcels into the anticyclones, Figure 3 displays the time-averaged three-dimensional wind fields for August 2011 from the ECMWF operational analysis, in terms of four horizontal cross sections through the greater Asian monsoon region of horizontal winds (vectors) and vertical motion (shaded). As expected for a region of active tropical convection, the spatial patterns of vertical motion are similar throughout the vertical column, with magnitudes weakening with altitude. Northeast India, the Bay of Bengal, and western Pacific, in particular, stand out as regions of strong uplift throughout the troposphere. Horizontal circulations are much stronger at upper levels (≤ 250 mbar; Figures 2a and 2b), where the anticyclone circulation is clearly evident at 15°N–45°N, 30°E–120°E. Both the horizontal and vertical components of the circulation are expected to influence the trajectory paths from the PBL to the ASM anticyclone. Since vertical motion is slow and typically downward outside of active convection, we expect that parcels reaching the anticyclone from the boundary layer to pass through the shaded regions of Figure 2. However, the strong upper level horizontal circulation also plays an important role by either guiding air parcels into or diverting them away from the anticyclone, favoring certain convective source locations over others.

Figure 3.

Vertical motion (shaded) and horizontal wind fields (vectors) for the ASM/anticyclone region at (a) 150 mbar, (b) 250 mbar, (c) 400 mbar, and (d) 600 mbar. Fields are scaled by their respective maximum values (displayed in each panel heading) with shading levels at 0.25 and 0.50 times the maximum upward pressure velocity (–ω). Values of vertical velocity are smoothed with a 5° cosine window. Pertinent features are as follows: (1) the regions of strongest uplift (NE India/Bay of Bengal and the western Pacific) are common to all levels, and (2) the horizontal winds associated with the anticyclone are strong at upper levels (p ≤ 250 mbar).

[18] Horizontal cross sections of the PBL-to-anticyclone pathway for the 200 mbar calculations are shown in Figure 4. Only trajectories with PBL sources contribute to these cross sections, and so, the number of contributing trajectories is identical for each panel. Furthermore, the shaded region in each panel accounts for 90%, and the black shading accounts for 20% of these trajectories. Thus, these maps are faithful and consistent representations of pathway cross sections. Figure 4a shows the locations of the trajectories on the σ = p/ps = 0.85 level, i.e., the geographical distribution of PBL sources. The next four successive panels show where trajectories with PBL sources first encountered the following levels: (b) σ = 0.75, (c) σ = 0.65, (d) 300 mbar, and (e) 220 mbar. These levels sample the troposphere while accommodating the large surface pressure variations over the Asian monsoon sector. Figure 4f shows the initial locations for trajectories with PBL sources; the spatial patterns in Figure 4f are very similar to the spatial pattern of the anticyclone location (Figure 2a) because trajectories with PBL encounters account for a large fraction (78%) of the overall trajectories. Throughout most of the journey from the PBL to the anticyclone, parcel locations are confined to a conduit that is centered over NE India, Nepal, and the southern Tibetan Plateau. It is only for pressures within about ~20 mbar of the 200 mbar anticyclone (Figure 4e) that the geographical distribution spreads to fill the anticyclone region. It is also noteworthy that the area of the region of boundary layer sources (Figure 4a) is somewhat larger than that of the conduit in the midtroposphere (Figures 4b–4d). That is, while sources of air for the anticyclone are gathered from a region covering much of southern Asia, most sources arise from and travel through a conduit that covers a relatively small area, finally spreading to fill the anticyclone over a short vertical distance (~20 mbar; see the exaggerated schematic in Figure 5). This schematic representation of the pathway to the anticyclone resembles the monsoonal flow into the anticyclone described by Park et al. [2009; their Figure 14].

Figure 4.

Trajectory pathways from the PBL to 200 mbar anticyclone using forcing fields from the ECMWF operational analysis smoothed to 1° resolution. Shown are first-encounter locations, as a function of pressure, for trajectories initiated within the 200 mbar anticyclone that encounter the PBL within 30 d. (a) Geographical distribution of boundary layer encounters. (b–e) Geographical distribution of those trajectories as they cross the levels: (b) σ = 0.75 (corresponding to pressures p = 450–750 mbar for surface pressure ps = 600–1000 mbar, respectively), (c) σ = 0.65 (p = 390–650 mbar), (d) 300 mbar, and (e) 220 mbar. (f) Initial positions at 200 mbar of trajectories with PBL sources. Contour levels were chosen so that 90% of the trajectories are included within the shaded region and 20% within the black shading. Values are smoothed with a 5° cosine window. Note that the pathways are confined to a conduit centered over southern Tibet throughout the midtroposphere (σ = 0.75 to p = 300 mbar).

Figure 5.

Schematic illustrating the conduit from the boundary layer to the anticyclone. Air parcels are gathered from a wide area of the PBL and then confined to a narrower region as they traverse the midtroposphere. At 20–50 mbar below the anticyclone, they spread to fill the areal extent of the anticyclone.

[19] Pathways to the 100 mbar anticyclone (Figure 6) conform to the same schematic description as the 200 mbar anticyclone. The 100 mbar anticyclone draws on PBL sources (Figure 6a) from nearly the same locations that the 200 mbar anticyclone draws from (Figure 4a). Parcels are then channeled through a conduit centered over NE India and southern Tibet (Figures 6b–6d) and finally spread to fill the 100 mbar anticyclone in the last 50 mbar of the journey (Figures 6e and 6f). A notable exception to the south Asian conduit is a path to the 100 mbar anticyclone in the western Pacific followed by parcels detrained from tropical cyclones. Another interesting feature is that parcels transported from the PBL to the 100 mbar anticyclone are still confined to the mid-tropospheric conduit at 200 mbar (Figure 6d). In fact, the preferred pathway from the 200 mbar anticyclone to the 100 mbar anticyclone for parcels in general (not just those with PBL sources; not shown) has a nearly identical spatial pattern as that shown in Figure 6d. Thus, it is the conduit identified here, not the anticyclone itself, that defines the most efficient pathway from lower levels to upper levels of the anticyclone.

Figure 6.

As in Figure 4, except for the 100 mbar anticyclone and that pressure levels are at (a) the PBL, (b) σ = 0.75, (c) σ = 0.65, (d) 200 mbar, (e) 150 mbar, and (f) 100 mbar. 60 d trajectories were used for these calculations so that the fraction of anticyclone trajectories that have boundary layer encounters (81%) matches approximately that fraction for the 200 mbar calculations (78%). Similar to results from the 200 mbar calculations, pathways are confined to a vertical conduit centered over southern Tibet throughout the midtroposphere.

[20] Figure 7 displays the location of the conduit to the 100 mbar anticyclone (σ = 0.65; blue contour lines) relative to the locations of the anticyclone (black contour lines) and to regions of strong upward vertical motion (shading). The orange contour identifies the Tibetan Plateau (August mean ps = 750 mbar) as a geographical reference. This figure reveals that the location of the conduit is neither centered within the core of the ASM anticyclone northwest of India nor within the strongest convective activity but between those two regions. This indicates that, to reach the anticyclone, it is important to have strong vertical motion that lofts air parcels directly into or upwind of the anticyclone. Over most of the anticyclone region, uplift is inadequate for parcels to reach 100 mbar efficiently, while over most of the regions of strong uplift, parcels are, on average, transported away from the anticyclone by upper level winds (vector arrows). This does not imply that these regions of strong uplift are not important sources of lower stratospheric air, only that they do not contribute to air within the anticyclone. In fact, previous studies [e.g., James et al., 2008; Bergman et al., 2012] have shown the Bay of Bengal to be an important convective source of air entering the stratosphere during boreal summer.

Figure 7.

The location of the midtropospheric conduit (σ = 0.65; blue contour lines) relative to the location of the 100 mbar anticyclone (black contour lines), to the tropospheric-averaged vertical motion field (–ω; shading), and to the upper-level horizontal winds (vector arrows) over the Asian monsoon/anticyclone region. Contours represent 0.25 and 0.5 of the maximum value for each field. Vertical motion is represented as a mass-weighted tropospheric average (top of the boundary layer to 100 mbar). For geographical reference, the orange contour outlines the Tibetan Plateau (August mean surface pressure ps = 750 mbar). Fields are smoothed with a 5° cosine window. Note that the midtropospheric conduit is located in close proximity to strong convection and to the center of the ASM anticyclone, but is precisely colocated with neither.

4 Regional Contributions: Robust Results of Sensitivity Experiments

[21] The location of the conduit has interesting consequences. For example, the Tibetan Plateau, which does not have exceptionally strong August mean uplift, nevertheless supports the highest concentration of PBL sources for the anticyclone (black shading in Figures 4a and 6a). In contrast, the Bay of Bengal, which has strong August mean vertical motion but is just south of the center of the conduit, is not an important source because that air is carried south of the anticyclone by upper level winds (note the southward component to the mean upper level winds over the Bay of Bengal in Figure 7). This section investigates these regional contributions more closely as a means to synthesize the information represented in Figures 4 and 6 and to quantify model sensitivities. In particular, we want to understand how unresolved components of the circulation and errors in the vertical motion fields are likely to affect our results. For this analysis, we use ensembles of trajectories forced by three different data sets: the ECMWF operational analysis, the GFS operational analysis, and the MERRA reanalysis and by three synthetically modified data sets (described in section 4.2). As in the previous section, trajectories are initialized every 1° longitude by 1° latitude (30°E-135°E; 15°N–45°N) and then filtered for geopotential height exceeding the threshold ΦAAC. However, trajectories are only calculated for 30 d (for bo th the 200 mbar and 100 mbar initializations) and initialized once daily (12 Z). The consistent use of 30 d trajectories reveals aspects of transport to 100 mbar that differ from transport to 200 mbar simply because the transport time to 100 mbar is longer in addition to allowing more computational flexibility to perform suites of sensitivity calculations. To determine the impact of the sampling strategy used here, we compared the regional contributions using once daily sampling to those using four times daily sampling with the reduced resolution ECMWF forcing fields. For all statistics cited in this section, data sampling made no difference to our results at the accuracy of the numbers quoted here (typically three significant figures).

4.1 Regional Contributions

[22] Figure 8 defines the regions, which are confined to one of two boxes that divide the ASM sector. The western box (Box 1; 0°–40°N; 60°E–105°E) is divided into three regions: the Tibetan Plateau (land with August mean surface pressure ps ≤ 750 mbar; bounded by the thick black contour in Figure 8), India/SE Asia (land with ps > 750 mbar), and the Indian Ocean. The eastern box (Box 2; 0°–40°N; 105°E–140°E) is divided into a land region (eastern China/Philippines) and an ocean region (western Pacific).

Figure 8.

Definitions for the five boundary layer source regions. The Tibetan Plateau is land areas within Box 1 with August 2011 mean surface pressure less than 750 mbar (thick contour), India/SE Asia is Box 1 land areas with ps > 750 mbar, eastern China/Philippines is Box 2 land areas, the Indian Ocean is oceanic areas in Box 1, and the western Pacific is oceanic areas in Box 2. The boundaries for Box 1 are 0°–40°N, 60°E–105°E; for Box 2, they are 0°–40°N, 105°E–140°E.

[23] Regional PBL contributions to air within the anticyclones are displayed with color bars in Figure 9; each bar represents results from a distinct set of forcing fields. These regional contributions are displayed as percentages of the PBL sources; i.e., they have been normalized by the total number of boundary layer encounters. The total number of PBL encounters is displayed as a fraction of the total trajectories to the right of each color bar. There are important features of these contributions that are robust among all data sets. For the 200 mbar calculations, approximately 78% of the PBL sources come from land (the range over all calculations is 76.2%–79.1%), 16% from ocean (14.9%–17.4%), with the rest from outside the specified regions. The land contributions are split approximately in half between the Tibetan Plateau (34.8%–41.3%) and India/SE Asia (34.1%–40.1%), with only a small contribution coming from eastern China/Philippines (2.9%–4.9%). Of the ocean regions, the Indian Ocean contributions (8.5%–12.6%) are nearly double the western Pacific contributions (4.5%–7.5%).

Figure 9.

The normalized regional contributions to young air within the Asian anticyclone from ensembles of trajectories calculated six different sets of forcing fields: the ECMWF operational analysis smoothed to 1° horizontal resolution (Low res. EC), the full-resolution ECMWF operational analysis (ECMWF), the GFS operational analysis (GFS), the MERRA reanalysis (MERRA), a combination of horizontal winds from MERRA and vertical velocity from ECMWF (Mixed MER-EC), and MERRA with vertical velocity adjusted to improve the linear relationship to TRMM precipitation (Adj. MERRA). Values on the left are calculated from 200 mbar anticyclone trajectories; those on the right are from 100 mbar anticyclone trajectories. The color bars represent the fractional contribution from each region to the total boundary layer source (i.e., the sum of the color bars is 100%); the fraction of 30 d trajectories with boundary layer sources are found in the column of numbers just to the right of the color bars. In general, the relative regional contributions are more consistent among experiments than are the total boundary layer encounters; values for 200 mbar trajectories are more consistent than those for 100 mbar, and continental sources dominate over oceanic sources.

[24] There are also robust features of the PBL-anticyclone transit times as illustrated in Figure 10, which displays the probability distribution functions (pdfs) of boundary layer encounters as a function of transit time for four of the regions and each of trajectory ensembles. We omit eastern China/Philippines, which has a noisier pdf but also contributes less to the boundary layer sources than the other regions. With the exception of trajectories sourced from the Indian Ocean (Figure 10c), the spread among the ensembles is small relative to the width of peaks in the pdfs. Thus, Figure 10 is able to resolve peaks in the pdfs for the Tibetan Plateau (~2-18 d at half maximum; Figure 10a) and for India/SE Asia (~3–22 d; Figure 10b), and to determine that, for the Indian Ocean (Figure 10c) and western Pacific (Figure 10d), the peaks are much broader though not fully resolved by 30 d trajectories. It is probably no coincidence that the two regions with well-defined peaks in the pdfs and shortest transit times (the Tibetan Plateau and India/SE Asia) are the two that are in close proximity to the mean location of the anticyclone (compare Figures 2a and 8) and convection over these regions can often inject boundary layer air directly into the anticyclone. The other two regions (Indian Ocean and western Pacific) are more remote, and so, their contributions rely on horizontal as well as vertical transport.

Figure 10.

Boundary-layer-to-anticyclone transit time pdfs from the 200 mbar trajectory ensembles for trajectories with PBL sources from (a) the Tibetan Plateau, (b) India/SE Asia, (c) the Indian Ocean, and (d) the western Pacific. The different lines in each panel correspond to calculations from the six different forcing fields: the ECMWF operational analysis smoothed to 1° horizontal resolution (Low res. EC; black dotted line), the full-resolution ECMWF operational analysis (ECMWF; black solid line), the GFS operational analysis (GFS; red solid line), the MERRA reanalysis (MERRA; blue solid line), a combination of horizontal winds from MERRA and vertical velocity from ECMWF (Mixed MER-EC; red dotted line), and MERRA with vertical velocity adjusted to improve the linear relationship to TRMM precipitation (Adj. MERRA; blue dotted line). Values are binned into 1 d increments, lightly smoothed with one pass of a 1-2-1 filter, and normalized to integrate to 1.0 over 30 d. Note the small spread for the different calculations particularly in Figures 10a, 10b, and 10d.

[25] It is also noteworthy that the spread among the different calculations in Figure 10 for short transit times (less than 5 d) is ordered, for the most part, according to the horizontal resolution of the forcing fields. That is, the pdf increases with time faster for calculations forced by the full-resolution ECMWF (solid black line) than for those forced by GFS (solid red) or the low-resolution ECMWF (dotted black) data, which increase faster than those forced by the MERRA reanalysis (solid blue). The difference between the calculations forced by the full-resolution ECMWF and those forced by the low-resolution ECMWF is particularly telling since horizontal resolution is the only difference between those data sources. As discussed in section 1, we expect unresolved variability, and thus horizontal resolution, to more strongly impact short transit times. Over several days, the effects of small-scale fluctuations will tend to cancel one another, although, admittedly, such cancellation is rarely complete (see, for example, discussions of Brownian motion and other forms of noise-induced drift in Garner [1985]).

[26] For the 100 mbar calculations, regional contributions (right side of Figure 9) are similar to those for the 200 mbar calculations but with shifts in the contributions that are uniform among all ensembles. There are two regions that have systematically larger contributions for 100 mbar than for 200 mbar: the Tibetan Plateau (on average, up from 37% at 200 mbar to 45% at 100 mbar) and western Pacific (from 6% to 11%). The other three regions have systematically smaller contributions at 100 mbar; India/SE Asia decreases from 37% to 30%, eastern China/Philippines from 4% to 2%, and Indian Ocean from 10% to 7%. The pdfs of transit times for these calculations (Figure 11) are noisier and contain less information than the 200 mbar calculations (Figure 10). Nevertheless, it is evident that the pdf peaks for the 100 mbar trajectories are both broader and encompass later times than the 200 mbar trajectories—as well they should, given that the additional vertical distance and weak vertical velocities above 200 mbar (Figure 3a) make transport times to 100 mbar longer than those to 200 mbar, and strong horizontal circulations above 200 mbar disperse air parcels, potentially exposing them to a wider range of vertical motion, which broadens the pdfs in Figure 11.

Figure 11.

As in Figure 10, but for the 100 mbar calculations for four different ensembles: the ECMWF operational analysis smoothed to 1° horizontal resolution (Low res. EC; black dotted line), the full-resolution ECMWF operational analysis (ECMWF; black solid line), the GFS operational analysis (GFS; red solid line), and the MERRA reanalysis (MERRA; blue solid line). Spread among the different calculations is wider than the 200 mbar calculations but nevertheless demonstrates that peaks in the pdfs are broader and occur 10–15 d later than the 200 mbar calculations.

4.2 The Role of Vertical Velocity

[27] There is one glaring discrepancy among the different trajectory calculations; while the normalized regional contributions and transit time probability distributions are fairly robust, the total boundary layer contributions (“Total” in Figure 9) are not. This is particularly true for the 100 mbar calculations, for which the total varies from 0.17 to 0.52. Spatial resolution likely plays a role in these discrepancies. For example, the 0.125° resolution operational analysis data set (ECMWF) finds PBL sources for 52% of the 100 mbar trajectories, the 1° resolution operational analyses (low-resolution ECMWF and GFS) find PBL sources for ~40% (38% and 35.8%, respectively), and the MERRA reanalysis finds PBL sources for only 17% of the trajectories. The horizontal resolution of MERRA reanalysis is only slightly coarser (1.25°) than GFS. However, vertical velocity in MERRA has been temporally averaged [Schoeberl and Dessler, 2011], which smoothes the horizontal dependence as well (see Figure 12d). The resolution dependence implied here, while based on a small number of experiments, is strengthened by the fact that the only difference between the full resolution and the 1° resolution ECMWF data sets is horizontal resolution.

Figure 12.

(a) Mean precipitation rate (mm/h) for August 2011 from TRMM together with the vertically averaged vertical motion –ω (mbar/h) from (b) ECMWF, (c) GFS, and (d) MERRA. Averages for vertical motion are weighted by the pressure thickness of each vertical layer, approximating a mass weighted average. All fields are smoothed in latitude and longitude with a 5° cosine window. Note the pattern similarity between (b) vertical motion from ECMWF and (a) precipitation from TRMM over a wide range of space scales and (d) the overly smoothed spatial patterns of MERRA vertical motion.

[28] For the 200 mbar calculations, total PBL encounters have a single outlier; the MERRA trajectories have PBL sources for only 62% of the trajectories, whereas all other calculations have PBL sources for 78.4%–82.5% of the trajectories. That is, the MERRA discrepancy is five times as large as the full range of values from the other calculations. That this discrepancy results primarily from differences in the vertical velocity is the focus of the remainder of this subsection. Figure 12 compares the geographical distributions of monthly tropospheric average vertical velocity (shown as the negative of pressure velocity ω) to that of precipitation rate from the Tropical Rainfall Measuring Mission (TRMM) Multisatellite Precipitation Analysis, version 3B42RT [Huffman et al., 2007]. We expect to find a strong relationship between precipitation and tropospheric vertical velocity, particularly in regions of tropical convection where heavy rainfall is associated with strong updrafts. Although we do not know what the correlation between precipitation and vertical velocity should be, the qualitative comparisons in Figure 12 favor ECMWF (Figure 12b) as having the most realistic vertical velocity field, as it exhibits favorable comparisons with precipitation (Figure 12a) on a wide range of spatial scales. Vertical velocity from GFS (Figure 12c) shares the same large-scale variability and even some small-scale features but has numerous ‘bull's-eyes” of strong vertical velocity that are not found in any of the other maps in Figure 12. Vertical velocity from MERRA is clearly much smoother than the other fields but has similar large-scale patterns. The apparent close relationship with precipitation exhibited by the ECMWF data in Figure 12 is verified by detailed linear comparisons of daily-mean tropospheric vertical velocity and precipitation; the linear correlation coefficient between ECMWF and TRMM is 0.75 (Figure 13b). That correlation coefficient is 0.66 for GFS (scatter plot not shown) and 0.57 for MERRA (Figure 13a).

Figure 13.

Scatter comparisons showing linear relationships between the following: (a) MERRA vertical motion and precipitation from TRMM, (b) ECMWF vertical motion and TRMM precipitation, (c) MERRA vertical motion and MERRA precipitation, and (d) adjusted MERRA vertical motion and TRMM precipitation. Data within the region 10°S–45°N, 25°E–135°E during August 2011 is binned within 5° latitude by 5° longitude by 1 d bins. The adjustment gives vertical motion from MERRA a linear relationship with TRMM precipitation that is similar to that between TRMM and ECMWF as well as the expected similarity to that between MERRA vertical motion and MERRA precipitation.

[29] To examine the role of vertical velocity for the discrepancy in total boundary layer encounters, we perform two experiments. In the first, we substitute vertical velocity from ECMWF into the MERRA data set. Such a substitution potentially introduces dynamical imbalances into the wind fields that violate mass continuity. Our goal here is not to create an improved MERRA data set, only to demonstrate the sensitivity of the boundary layer encounters to changes in vertical velocity. This change (Mixed MER-EC in Figure 9) raises the 30 d fraction of trajectories with PBL encounters to 0.77, nearly the same as the low-resolution ECMWF trajectories (0.78), while making relatively minor changes to the normalized regional contributions. That this simple substitution nearly eliminates the large discrepancy is a clear demonstration that the vertical motion fields are at the root of the problem. That MERRA is the sole outlier and seems to have the least realistic vertical motion fields of the analysis data sets used here is evidence for the case that the MERRA vertical motion fields are problematic. The next experiment essentially seals the case.

[30] The second experiment also adjusts the vertical velocity from MERRA. This adjustment is based on the expectation that precipitation and vertical velocity fields from MERRA are dynamically consistent with each other. This can be the case even for unrealistic precipitation fields because convective precipitation depends heavily on convective updrafts that also dominate vertical velocity fields. This expectation is supported by the fact that the linear relationship between vertical velocity from MERRA and precipitation from MERRA (Figure 13c) is similar to the linear relationship between vertical velocity from ECMWF and precipitation from TRMM (Figure 13b). What happens if we adjust the MERRA velocity to have the same linear relationship with TRMM that it has with MERRA precipitation? To make this adjustment, we first perform a linear regression of MERRA vertical velocity at all grid points onto MERRA precipitation at the corresponding latitude and longitude.

display math(3)

[31] ωMERRA is the pressure velocity from MERRA, α is the regression coefficient, PMERRA is the precipitation from MERRA, and ξ is the residual variability in ω once the component that is linearly related to precipitation is removed. The adjusted vertical velocity ωadj is then calculated by substituting TRMM precipitation PTRMM for PMERRA in equation (3).

display math(4)

[32] The resulting scatter plot between ωadj and PTRMM is displayed in Figure 13d. Note the close quantitative agreement between the tropospheric average ωadj-PTRMM relationship and the ωMERRA-PMERRA (Figure 13c) relationship that is imposed by the adjustment as well as the quantitative agreement with the ωECMWF-PTRMM (Figure 13b) relationship that is not imposed. This adjustment results in total PBL encounters and regional contributions (Adj. MERRA; Figure 9) that are, perhaps, strikingly similar to the Mixed MERRA-ECMWF case discussed above. The fact that this adjustment, based on the linear relationship between MERRA vertical velocity and MERRA precipitation and applied to TRMM precipitation, is so similar to swapping ECMWF vertical velocity into the MERRA data is strong supporting evidence for two contentions: (1) vertical velocity is largely responsible for the discrepancies between the trajectories from MERRA and those from ECMWF and GFS, and (2) the trajectories from ECMWF are likely to be more realistic than those from either MERRA or GFS.

4.3 Diabatic vs. Kinematic Trajectories

[33] We conclude with a brief discussion of the difference between kinematic and diabatic trajectories for the applications discussed in this paper. Figure 14 displays the normalized regional contributions for diabatic trajectories calculated from MERRA horizontal winds and total diabatic heating rates. (Note: Diabatic heating rates are not available for the ECMWF and GFS operational analyses from our data sources.) At 200 mbar, the regional contributions as well as the total PBL encounters are similar to those of the kinematic MERRA trajectories (Figure 9), although the diabatic calculations have more PBL encounters over land and fewer over ocean. The PBL encounters for 100 mbar diabatic calculations, however, have little in common with any of the kinematic calculations. In fact, PBL encounters using MERRA diabatic heating rates are very rare—less than 1% of the trajectories. Further investigation reveals that this is due to the fact that cooling (and thus downward motion) dominates the MERRA diabatic heating rates at 150 mbar. This cooling is neither consistent with radiative transfer calculations [e.g., Yang et al., 2010] nor with the observed convective cloud activity in the upper troposphere [e.g., Fu et al., 2006; Devasthale and Fueglistaler, 2010]. Thus, the discrepancy between the diabatic and kinematic calculations is probably less of an indication of a difference between two trajectory approaches than it is an indictment of the MERRA diabatic heating rates in the boreal summer TTL. The lack of convective activity above 200 mbar implied by the MERRA heating rates supports our speculative interpretation of the resolution dependence of PBL encounters for the 100 mbar calculations identified in section section 4.2: very deep, small-scale convective cells that are typically not well represented in low-resolution data sets [cf. Arakawa, 2004] are important for the transport of air from the PBL to the 100 mbar anticyclone on time scales of 30 d and less.

Figure 14.

As in Figure 9, regional contributions to boundary layer sources for the 200 mbar anticyclone (top) and 100 mbar anticyclone (bottom) from diabatic trajectories calculated using total diabatic heating rates from MERRA. At 200 mbar, the values are consistent with the kinematic calculations (“MERRA” in Figure 9). At 100 mbar, the discrepancies with kinematic calculations are large.

5 Conclusions

[34] We investigated the transport of air from the boundary layer into the ASM anticyclone with back trajectories initiated within the anticyclone at 100 mbar and 200 mbar during August 2011. By analyzing the locations of trajectories as a function of vertical level, a well-defined transport pathway is revealed; PBL sources for the anticyclone are gathered from a region covering much of southern Asia, traverse the midtroposphere through a vertical conduit centered over NE India, Nepal, and southern Tibet, and finally spread to fill the anticyclone over a short vertical distance (20–50 mbar). The vertical conduit has two important properties: (1) The pathways to the 100 mbar anticyclone are nearly as geographically confined at 200 mbar as they are at midtropospheric levels. Thus, it is the conduit identified here, not the anticyclone itself, that defines the most efficient pathway from lower to upper levels of the anticyclone. While the anticyclone is a coherent dynamical feature of the upper troposphere and lower stratosphere, it does not by itself define a transport pipeline through that region. (2) The conduit is in close proximity to both the mean location of the anticyclone and the mean location of strong vertical motion but is colocated with neither, indicating that, in order to enter the anticyclone from the boundary layer, air parcels require strong vertical motion that is located within or directly upwind of the anticyclone. While the conduit represents the most efficient pathway from the PBL to the anticyclone, it should be noted that our experiments do not determine the efficiency with which the conduit transports air into the anticyclone. In fact, the air that ultimately enters the ASM anticyclone from the PBL in regions identified in Figures 4a and 6a might only be a small fraction of the air that is convectively lifted within the conduit. That efficiency of the conduit for transporting air to the AAC will be best quantified with forward transport calculations initialized within the source region boundary layer.

[35] To synthesize and quantify the PBL source information for more rigorous analysis, we calculated regional contributions to air within the anticyclone by boundary layer sources from five regions: the Tibetan Plateau, India/SE Asia, eastern China, the Indian Ocean, and the western Pacific. For this analysis, we limited the trajectories to 30 d and performed calculations using wind fields from multiple data sets. Three principal data sets were used: the ECMWF operational analysis, the NCEP-GFS operational analysis, and NASA-MERRA reanalysis. Additional calculations were performed to explore the roles of vertical motion and of horizontal resolution for discrepancies among these calculations. Important features of the regional contributions, including their transit times, were found to be robust among all kinematic calculations, promoting confidence in these results. Boundary layer sources for the anticyclone are primarily from the Tibetan Plateau and India/SE Asia (a combined 70%–80%) at both the 200 mbar and 100 mbar levels, with minor contributions by the western Pacific and Indian Ocean (a combined 15%–20%). The Tibetan Plateau and the western Pacific are both more important sources of “young” air (PBL-to-anticyclone transit times < 30 d) for the anticyclone at 100 mbar than they are at 200 mbar. The different trajectory calculations also have similar transit time pdfs. Transit times are shorter for the Tibetan Plateau and India/SE Asia than for western Pacific and Indian Ocean—presumably due to the geographical proximity of the former two regions to the ASM anticyclone.

[36] While the different calculations agree on the relative contributions by the regions and even on the regional transit times, there are substantial differences in the total fraction of young air within among the data sets. At 200 mbar, MERRA is the clear outlier (62% young air compared to 83% for ECMWF and 81% for GFS). This discrepancy is attributed to the unrealistically smooth vertical motion field in MERRA. When the MERRA vertical motion is adjusted to have a more realistic relationship with TRMM precipitation, the discrepancy is effectively removed. Large discrepancies in the total number of boundary layer sources for the 100 mbar anticyclone are not resolved via the relationship between vertical motion and precipitation. In addition, 100 mbar calculations are more sensitive to the horizontal resolution of the wind data than the 200 mbar calculations. These sensitivities are particularly interesting in light of the fact that tropical convection is known to have large variability at small horizontal scales and potentially important strong convective updrafts are not resolved by analysis data sets. That the 200 mbar calculations are not very sensitive to changes of horizontal resolution implies that transport on 30 d time scales is well represented by the aggregate vertical motion of ensembles of convection. For certain, our calculations do not resolve the explicit transport by strong small-scale updrafts, but it seems likely that over a 30 d period, such transport either is accounted for by the averaging effect of resolved scales or is unimportant. This is not the case for the 100 mbar calculations for which the averaging effect of low-resolution data sets makes vertical transport to 100 mbar too slow to accurately represent actual transport over 30 d. This speculation highlighting the role of unresolved deep convective updrafts for feeding the 100 mbar ASM anticyclone is also supported by diabatic calculations using MERRA. The 200 mbar calculations using MERRA heating rates agree well with the MERRA kinematic calculations. However, young air at 100 mbar is nearly absent in the diabatic calculations due to the (unrealistically) low diabatic heating rates at 150 mbar in MERRA, which in turn indicates a lack of high clouds. If the lack of convection and high clouds in the MERRA reanalysis data plays as important of a role in the discrepancy of trajectory results as it seems, then convection and high clouds must be important for transport from 200 mbar to 100 mbar.

[37] Our results touch on an issue that has been discussed by several investigations in the recent past, the importance of convection over the Tibetan Plateau for transporting air into the lower stratosphere. At a superficial level, our results agree with studies that support an important role for the Tibetan Plateau [e.g., Fu et al., 2006; Wright et al., 2011] and disagree with those that do not [e.g., Park et al., 2009; Devasthale and Fueglistaler, 2010; James et al., 2008; Chen et al., 2012]. However, when diagnosing transport, details of the analysis are crucial to the interpretation of the result. If chemical tracers are used for the diagnosis, then the results depend on the distribution of tracer sources as well as the properties of the dynamical transport. Furthermore, backward transport calculations such as ours can diagnose the sources of air for a specific target region but are poor indicators of the efficiency with which air from the sources feed the target. The reverse is true for forward calculations. In addition, we only diagnose the contributions to air within the anticyclone, which, while important, is not the sole pathway through the TTL. Thus, it is not actually possible to diagnose the importance of convection over the Tibetan Plateau for transport into the lower stratosphere from our results. In light of these considerations, we find no contradictions between our results and any of the aforementioned studies.


[38] This work benefitted from helpful conversations with and comments on early versions of the manuscript from C. Homeyer, S. Fueglistaler, H. Garny, M. Park, L. Pfister, W. Randel, and two anonymous reviewers. J. Bergman. F. Fierli, and E. Jensen were visitors at the Atmospheric Chemistry Division of NCAR during the execution of this study. The National Center for Atmospheric Research is operated by the University Corporation for Atmospheric Research, under sponsorship of the National Science Foundation.