A particular pattern in the global distribution of aerosols, characterized by high concentrations of ultrafine particles (mostly nucleation- and Aitken-mode particles) in the tropical upper troposphere (UT), between ≈ 9–12 km altitude, has been documented over the last three decades e.g.,. These particles have been attributed to nucleation occurring in the outflow of deep convective clouds, associated with the transport of precursor gases (sulfuric acid and/or organic vapors) from the atmospheric boundary layer .

Enhanced aerosol content in the UT has important implications for Earth's radiation budget and may serve as a source of cloud condensation nuclei (CCN) following particle growth and downward transport . For this reason, accurately representing these aerosols is necessary in weather and climate simulations. Attempts to simulate particle production in the UT have been reported by, for instance, , , , , and . However, it remains challenging to emulate the UT aerosol enhancement within the tropics because of the difficulties in adequately representing sources and sinks of precursor gases, chemical reactions leading to new particle formation, and transport mechanisms. Such representation depends largely on physical parameterizations, including surface emissions, cloud microphysics, convection, and aerosol/gas chemistry, as well as on dynamical feedbacks.

Specifically, transport from the tropical UT may follow different pathways. While transient phenomena – such as low- and high-pressure systems at the synoptic scale, and even mesoscale convective systems – can induce significant short-term variability, the general circulation patterns are expected to dominate in the long term. From a general circulation perspective, part of the UT air may be transported poleward within the Hadley cell and descend in the subtropics, or ascend further into the lower stratosphere above equatorial regions, where it may then be transported poleward and descend at high latitudes within the stratospheric (Brewer–Dobson) circulation on timescales of over 2 years . These mechanisms are consistent with multi-year statistics of aerosol observations showing a pronounced latitudinal gradient of aerosol concentration in the mid-to-upper troposphere, decreasing poleward from the tropics e.g.,. Similarly, the modeling study of suggests particle transport and growth from the tropical UT to the subtropical free troposphere. also showed that particle nucleation in the free troposphere and UT led a nearly zonal distribution at ground level, with maximum concentrations near the tropics likely associated with the downward branch of the Hadley cell.

Several observational studies have more directly linked downward transport to aerosol populations in the atmospheric boundary layer, with some more recent work suggesting a possible contribution from UT-origin particles e.g.,. However, observations typically sample the atmospheric state only intermittently in space and/or time, making it difficult to disentangle the transport history of sampled air masses and to distinguish among possible source levels. In other words, it remains unclear over what altitude range CCN budgets are dominated by UT-nucleated particles that grow during descent, and at what levels boundary-layer or free-tropospheric sources become dominant.

For a limited area over the unperturbed Amazon rain forest, the simulations of found net downward transport over a wide altitude range, from the UT to the lower troposphere. However, the underlying mechanisms behind this downward transport and their associated timescales are not yet thoroughly understood. For example, the relative roles of isolated and organized convection versus synoptic- or large-scale subsidence remain unexplored. Studies suggest that the connection between UT aerosol formation and lower-level aerosol populations is unlikely to occur through a single direct transport mechanism. For instance, the idealized simulations of showed that Amazonian convective downdrafts can efficiently transport tracers from the middle troposphere, around 5 km, to the boundary layer, whereas direct downward transport from upper-tropospheric levels to the boundary layer within one convective event was not evident in their simulations. Similarly, found that particles nucleated in the UT in a 4 km-resolution regional Amazonian domain were rarely transported into the boundary layer over several days. Instead, most boundary-layer CCN arriving from higher levels originated outside the region. Because these studies focused on relatively short timescales and limited-area domains, the question remains whether, and how, UT aerosols might be transported downward on longer timescales or across broader spatial scales.

Here, we address this gap by combining multi-year global simulations with idealized tracers initialized in the UT. Our focus is on the large-scale pathways and timescales by which UT air is transported into the mid-troposphere, where subsequent smaller-scale processes may become important. To achieve such extended temporal and spatial coverage, we necessarily use a coarser resolution than most of the aforementioned studies, yet we hypothesize that this approach allows us to isolate large-scale transport pathways and capture the long-term circulation patterns capable of carrying UT air – and the aerosols it contains – downward into the mid-troposphere.

Analyzing typical transport timescales and pathways provides a first step toward characterizing the fate of freshly-nucleated UT particles. To separate transport effects from uncertainties in aerosol formation and growth, we employ inert, passive tracers under an idealized initialization scheme, assuming a constant size distribution. To estimate particle survival rates, besides accounting for intrinsic dilution caused by numerical and parameterized mixing, we incorporate removal processes via wet scavenging, sedimentation, and dry deposition. The uncertainties associated with neglecting particle growth are assessed through a sensitivity test using a larger mean size and offline estimates of the impact of coagulation.

The design of the numerical experiments and the analysis methods are detailed in Sect. . Results, including tracer transport timescales and spatial patterns of tracer number mixing ratio, are presented in Sect. . A discussion of the relevance and applicability of these results is provided in Sect. , followed by concluding remarks in Sect. .

We employed the EMAC model , which is a combination of the ECHAM5 atmospheric general circulation model with the Modular Earth Submodel System MESSy;.

In the vertical dimension, equations are solved using finite differences , based on a hybrid pressure-based terrain-following coordinate with 90 levels and top at 1 Pa. A T106 triangular spectral truncation was chosen for the spherical harmonics dynamical core, with a corresponding 1.125° resolution (≈ 125 km at the Equator) quadratic Gaussian grid for the evaluation of advection terms and subgrid scale parameterizations. Tracer advection follows the flux-form semi-Lagrangian scheme of .

A semi-implicit scheme is used for time integration, with implicitness factors βDT=0.8 for divergence, temperature and surface pressure and βZQ=1 for vorticity, and with reference temperature and surface pressure of 300 K and 800 hPa, respectively . A time filter is used to limit the growth of spurious computational modes similar to , with a time filtering coefficient ϵ=0.1.

Two simulations were performed, starting on 1 January 1998: an 11-year simulation containing the age-of-air and continuously forced constant tracers, and a 2.5-year simulation containing the staggered-tracer sets; the tracer forcing configurations are summarized in Table  and described below. After excluding the first year as spin-up, the longer simulation provided up to 10 years for the analysis of the age-of-air and constant tracers, whereas the shorter simulation provided sufficient time to diagnose early-arrival timescales using the staggered tracers.

An integration time step length of 240 s was employed for the 2.5-year simulation and the first 2 years and two months of the 11-year simulation. At later times in the 11-year simulations, the integration time step length was reduced to 60 s to prevent computational instabilities.

Radiation processes and surface temperature parameterization follow the original ECHAM5 approaches, the former modified according to . The scheme of is employed to represent land-atmosphere exchange and vertical diffusion. Orographic gravity waves and low level drag are represented by the scheme of and , as in ECHAM5. Cloud microphysics are parameterized following , with cloud cover calculations following and . A mass flux convection scheme is applied , with the closure assumptions of .

Initial conditions were derived from ERA5, the fifth generation ECMWF (European Center for Medium-Range Weather Forecasts) reanalysis . The simulations were not dynamically nudged to reanalysis fields after initialization; instead, the model evolved freely from the initial state, forced only by prescribed surface boundary conditions. The sea surface temperature and sea ice concentration data employed as surface boundary conditions are based on a climatology computed as the multi-year monthly mean over 2000–2019, using output from the Max Planck Institute for Meteorology Earth System Model version 1.2 (MPI-ESM1.2) model prepared for the Coupled Model Intercomparison Project (CMIP6) .

To analyze transport pathways and associated timescales from the tropical UT, we introduced 19 idealized tracers (Table ) via MESSy's submodel PTRAC . The tracers were initialized and forced in the tropical UT between 200–300 hPa, except in some cases (suffix “_higher” in the tracer name), in which the forcing region was shifted to 150–250 hPa to evaluate the impact of uncertainties in the height of the tracer layer. Tracer initialization and forcing used the MESSy submodels IMPORT_GRID and TNUDGE . Different longitude intervals were defined for the forcing region: all over the Tropics (prefix “tropical_” in the tracer names), restricted to South American longitudes and centered in the Amazon (prefix “amazon_”), over the Sub-Saharan Africa (prefix “africa_”), and over the Maritime Continent (prefix “mar-cont_”). By considering these different forcing regions, we aim to understand the potential implications of longitudinal gradients in the aerosol source region for the transport pathways and timescales. The focus on continental regions is due to their higher potential for UT secondary aerosol nucleation compared to maritime regions, given the strong coupling between vegetation emissions and convective uplift.

A set of 5 tracers followed an “age of air” approach (“aoa” string in the tracer names in Table ), in which the tracer number mixing ratio Naoa varied quasi-linearly (i.e., step wise) with time within the forcing region and were subject to transport outside of it (Eq. ). The transport operator T includes resolved advection, as well as parameterized turbulent diffusion and convection. 1Naoa(t,P,ϕ,λ)=N∗max⁡0,t-1+e-2ϕϕ02,ifϕ∈[-ϕ0,ϕ0],λ∈[λmin⁡,λmax⁡],P∈[Pmin⁡,Pmax⁡].TNaoa,otherwise,

where t is the time in days; ϕ and λ are the latitude and longitude, respectively, in degrees; P is the pressure; N∗=2.08×1016 kg⁻¹ and ϕ0=20°.

The age of air is defined as the time elapsed since a parcel of air was last in contact with a specific source region. Due to mixing, the age at a given location is better described statistically by the “age spectrum”, denoted by Γ(τ), which represents the fraction of air mass with transit time τ. The mean age of air corresponds to the first moment of this distribution: 2〈τ〉=∫0∞τΓ(τ)dτ, representing the average transit time from the source to the target location .

In practice, 〈τ〉 can be estimated using a tracer that is uniformly distributed within the source region and increases linearly in time, with no sources or sinks elsewhere. Outside the source, the tracer is passively transported. At a given location, 〈τ〉 is estimated by subtracting the local tracer value from the value in the source region and dividing by the known rate of increase in the source region (i.e., the linear slope). This approach is described by and references therein. Note that, for tracers defined according to Eq. (), Naoa is not uniform within the forcing region –it is maximum at the Equator and decreases toward the maximum value on the previous day plus e-2 at ϕ=±20°. Smoothing the tracer spatial distribution in this way alleviates issues associated with advection near sharp gradients. However, even in the presence of spatial inhomogeneities, this approach remains valid for deriving 〈τ〉 with resolution of 1 d, given that the tracer values across the forcing region do not overlap in time.

To analyze the impact of size-dependent aerosol sinks on the transport patterns, the number mixing ratios of 6 tracers were kept constant within the forcing region throughout the simulation (“constant” string in the tracer names in Table ), mimicking a continuously-replenished aerosol reservoir: 3Ncnt(t,P,ϕ,λ)=N∗e-2ϕϕ02,ifϕ∈[-ϕ0,ϕ0],λ∈[λmin⁡,λmax⁡],P∈[Pmin⁡,Pmax⁡],t>t0.T(Ncnt)+F(Ncnt),otherwise, where t0=0 for the constant tracers discussed here, and F represents the sources/sinks associated with sedimentation, wet scavenging and dry deposition. According to Eq. (), the average of Ncnt within the forcing region (Nref) is ≈ 1.25×1016 kg⁻¹.

Some of the constant tracers were subject to sedimentation, wet scavenging, and dry deposition (i.e., F(Ncnt)≠0, suffix “with_ssdd”), where “ssdd” denotes the inclusion of these subgrid-scale processes as parameterized by and , via the MESSy submodels SEDI, SCAV, and DDEP, respectively. Here, sedimentation refers to gravitational settling through the atmospheric column, whereas dry deposition refers to removal at the surface. In contrast, other tracers were only transported (i.e., F(Ncnt)=0, suffix “no_ssdd”).

In order to apply these parameterizations, the tracers were assumed to represent aerosol particles distributed according to a log-normal function, with mean radius rm=20 nm and geometric standard deviation equal to 1.6. The value of rm was increased to 100 nm for one of the constant tracers (“bigger” string in the tracer names), to estimate the impact of uncertainties in the particle size on the parameterized sinks.

Adjusting the diagnostic framework to consider not only 〈τ〉 – which reflects the mass-weighted average over the full tracer age distribution Γ(τ) – but also transit times in the left tail of the distribution enables the characterization of faster, more immediate impacts on the target region. To diagnose such timescales, we introduced eight sets of 12 “staggered” tracers Nstgrd, defined identically to Ncnt in Eq. (), except that each tracer in a set has a different source start time t0=t0,m, corresponding to the first day of month m in 1999. After t0,m, the tracer is continuously maintained at the prescribed value inside the forcing region until the end of the 2.5-year simulation. These sets of tracers are denoted by the “_stgrd” suffix in the tracer names in Table : one set for each forcing region, and four additional sets corresponding to sensitivity tests for the Amazon source region. These include “_bigger” and “_higher” (as described above), and “_noconv”/“_nodiff” tracers for which parameterized convection or turbulent diffusion, respectively, were disabled.

Specifically, the elapsed time between tracer release and the moment a given fraction of the forcing-region mean number mixing ratio Nref is reached at a location can serve as a proxy for the influence of small amounts of material transported over relatively short timescales. We define this metric as the time-to-threshold tp, where p represents the chosen fractional threshold of Nref. Since the staggered tracers include not only transport but also atmospheric sinks such as wet scavenging, dry deposition, and sedimentation, the time-to-threshold metric reflects not the pure transport-derived age spectrum, but a modified or effective response that is shaped by both transport and removal processes. Conceptually, this response can be viewed as sampling a filtered age spectrum Γ∗(τ)=Γ(τ)S(τ), where S(τ) represents a survival function accounting for tracer loss as a function of age. The diagnostic can then be written as 4∫0tpΓ∗(τ)dτ=p. As such, the resulting timescales represent a convolved measure of both transit and retention, rather than transport alone. This formulation provides a measure of how rapidly a given fraction of the surviving tracer signal has arrived from the source, and captures the timing of fast transport components. Together, 〈τ〉 and tp provide complementary perspectives on downward transport. In the analysis, tp represents the arithmetic mean of the twelve time-to-threshold monthly estimates. This reduces the dependence of the diagnosed timescale on the arbitrary choice of release month.

To investigate downward transport from the UT, we take the 500 hPa level as a representative reference for the mid-troposphere. Figure  provides an overview of the simulations, showing time series of the tracer mixing ratio Ncnt and age of air 〈τ〉 over the full 11-year period. The values are averaged over horizontal areas centered on the forcing regions but extending 10° beyond them in each direction (larger red rectangles in Fig. ), at 500 hPa. The time series evidence a fast increase in 〈τ〉 (Fig. a, b) at the beginning of the simulation, as the target regions progressively fill with air originating from the forcing region. Note that grid points not yet reached by air from the forcing region are treated as missing values and are therefore excluded from the averages.

For the tropical tracers (Fig. a), the time series stabilize relatively quickly, reaching a plateau shortly after one year. In contrast, the regional tracers (Fig. b) require considerably more time to reach equilibrium (≈ 4 years). This difference reflects the source-region dependence of the mean-age diagnostic. All tracers are affected by dilution through mixing with air outside the forcing region. However, this effect is stronger for the regional tracers because their source regions occupy a much smaller fraction of the tropical UT. For the tropical tracer, air arriving at 500 hPa from any tropical longitude contributes to the source-tagged signal. For the regional tracers, however, the receptor regions also receive air from outside the prescribed source longitudes, which further dilutes the tracer signal and broadens the effective transit-time distribution. Typical values of 〈τ〉 derived from the tropical tracer are approximately 40–45 d, increasing to approximately 55–65 d when considering the higher source region. In contrast, 〈τ〉 derived from the regional tracers is approximately 260 d (Fig. a, b). This larger value should therefore be interpreted as a consequence of the smaller source-region area, enhanced dilution, mixing, and source-receptor geometry.

Consistent with the age of air, the 500 hPa area-averaged values of Ncnt from the tropics-wide source take approximately one year to reach a plateau (Fig. c). For this tracer, Ncnt/Nref plateaus at approximately 0.73, indicating a 27 % reduction due to dilution during downward transport. In contrast, when dry deposition, wet scavenging and sedimentation are applied, Ncnt/Nref stabilizes much faster, albeit at a lower value (≈ 0.15–0.25 for the tropical tracer and ≈ 0.05–0.15 for the regional ones), owing to the sink effect of these processes. Further analysis (not shown) indicated that wet scavenging is by the far the dominant sink in play. Little sensitivity to the role of wet scavenging in relation to variations in rm over the 20–100 nm range is evident by comparing the default and bigger tropical tracer with sinks in Fig. c. A bias in Ncnt/Nref of around ≈ 0.05 for the tropical tracer with sinks, and ≈ 0.025 for the regional tracers with sinks, before and after the year 2002 is evident in Fig. c, d. Given that this discontinuity is only evident in the time series of tracers with sinks, and that the effects of dry deposition and sedimentation are negligible (as noted above), the bias is likely associated with the wet scavenging parameterization and its sensitivity to the integration time step, which was reduced after 2002.

The results presented in Sect.  provide a comprehensive view of the timescales, spatial pathways, and dilution effects associated with the vertical transport of tracers from tropical, UT source regions to the mid-troposphere, analyzed from an Eulerian global-model perspective.

Several robust patterns emerge from the analysis, with important implications for our understanding of aerosol transport from the tropical UT to the mid-troposphere. First, the distinction between tropical-wide and regional tracer sources highlights the sensitivity of the mean-age diagnostic to the spatial extent of the emission region (Fig. ). The regional tracers exhibit significantly longer mean ages of air (〈τ〉≈260 d) than their tropical counterpart (〈τ〉≈45 d). This difference should not be interpreted as a proportional weakening of vertical transport, but rather as a consequence of the source-region dependence of 〈τ〉. For more geographically constrained sources, the diagnosed mean age is more strongly affected by mixing with air that has not recently been in contact with the source region (i.e., large τ components), which dilutes the source-tagged signal. A parallel can be drawn to previous variability-lifetime studies, which show that tracer spatial variability – and, by analogy, average age – is sensitive not only to residence time but also to the spatial distribution of sources and sinks e.g.,. In general, enhanced widening of the age spectrum in the troposphere is expected, as it is linked to strong mixing in this layer. This contrasts with stratospheric conditions, where spatial variability is more strongly governed by residence time , and the age-of-air approach is more straightforwardly applicable e.g.,, owing to the strong stratification that limits vertical exchanges compared to the troposphere.

In the context analyzed here, the time-to-threshold diagnostic, particularly at the 10 % level, provides a valuable complement to the mean age of air by capturing the onset of tracer influence in a given target region, effectively emphasizing the fastest-arriving component of the sink-filtered transport response. The broadly similar spatial patterns between these two diagnostics (Figs.  and ) affirm the robustness of the large-scale subsidence pathways, while the shorter timescales (often < 7 d) highlight the potential for relatively rapid vertical connections to the mid-troposphere.

Although not explicitly shown here, these results imply that aerosol transport from the UT to even lower levels – such as the atmospheric boundary layer, where low and convective clouds typically form – would require significantly longer than one week if governed solely by resolved-scale advection. Therefore, while reported consistent downward fluxes from the Amazonian UT to the boundary layer during their week-long simulations, our results suggest that such cross-troposphere connections would be difficult to achieve within that short timescale if governed solely by resolved-scale advection. Faster mechanisms – such as convective downdrafts, whose specific role in this context has not been fully elucidated – may therefore play an important role.

The spatial structure of the constant-tracer number mixing ratios Ncnt under the combined influence of transport and sinks (Fig. ) reinforces the conclusions. Regional tracers with constant sources and sinks dominated by wet scavenging exhibit peak mid-tropospheric number mixing ratios just east of their source regions – and slightly southwest in the case of the Maritime Continent tracer – aligning with the pattern of minimum transport times. This indicates that descent is not purely vertical, but occurs along slightly tilted pathways shaped by horizontal advection during subsidence. The lower-level regions of strongest influence remain within or near the UT source regions, but are displaced relative to the source centers. This is qualitatively consistent with the short-term simulations of , which showed that aerosols entering the boundary layer over the Amazon often originated outside the region. Nonetheless, the duration of their simulation represents an important constraint on the generality of their results. Our findings indicate that, despite the tilted descent path over the Amazon, significant downward transport may still occur within the Amazon region – i.e., within the domain simulated by – provided the transport timescale exceeds one week.

The resulting largely zonal distribution of tracer mixing ratios and transit times found here is broadly consistent with the results of , except that the tropical maxima reported in their study are not evident in our simulations. This difference likely arises from our constraint of the source region to latitudes between ±20°, whereas in , the nucleation rate evolved freely in space, depending solely on precursor availability and environmental conditions.

The reduction in number mixing ratios from the UT to the mid-troposphere in our simulations is substantial, but values of Ncnt/Nref≈0.1–0.15 remain meaningful and potentially important in the absence of other aerosol sources at lower altitudes. Notably, this fractional abundance is comparable to the contribution of secondary organic aerosols nucleated in the boundary layer relative to those formed in the upper troposphere in the numerical study of .

The quantitative values reported here are subject to uncertainties associated with the parameterized transport and removal processes in EMAC. Among the removal processes, wet scavenging is the dominant sink for the tracers with removal and therefore likely represents the largest parameterization-related uncertainty for the absolute values of Ncnt/Nref. Increasing the mean particle radius to 100 nm leads to a ≈ 22 % increase in descent time, mostly due to enhanced wet scavenging, while dry deposition and sedimentation play a smaller role for the particle sizes considered here.

Parameterized convection and turbulent diffusion can affect the vertical redistribution of tracers, but the sensitivity experiments in which these processes are disabled suggest a modest impact on t10% at 500 hPa. Neglecting convective transport increases descent time by ≈ 10 %, while removing turbulent diffusion yields a minor delay of only ≈ 1 % (Fig. b). Among all sensitivity experiments, the height of the tracer release emerges as the dominant factor controlling descent times. Raising the source region by 50 hPa (from 200–300 to 150–250 hPa; approximately 1–1.5 km higher) increases t10% by ≈ 19 d on average, which is large compared with the baseline value of t10% ≈ 25 d for descent from 300 to 500 hPa (≈ 3–4 km). This contrast points to nonlinear vertical transport behavior, presumably governed by factors such as the vertical profile of stratification and radiative cooling rates. In contrast, the influence of particle size, convection, and turbulence in our simulations is modest.

These results indicate that large-scale advection, together with implicit diffusivity, dominates the mean transport from the UT to the mid-troposphere at 500 hPa, at least on regional and climatological scales. Nevertheless, the sensitivity experiments should be interpreted as idealized on/off tests rather than a complete uncertainty quantification of the schemes. Thus, the absolute tracer abundances and threshold times should be regarded as model-dependent, whereas the broader spatial patterns and the dominance of large-scale advection in the mid-tropospheric transport appear more robust within the set of experiments considered here.

This study provides a detailed assessment of vertical transport timescales and spatial distribution of aerosol tracers released from tropical and tropical-continental UT source regions based on large-scale transport mechanisms. The key findings are:

Transport timescales are highly sensitive to the extent of the source region. Regional continental tracers exhibit mean air ages of ≈ 260 d at 500 hPa, compared to ≈ 45 d for tropical-wide sources. This difference reflects the source-region dependence of the mean-age diagnostic, including the effects of mixing, dilution, and source-receptor geometry for spatially confined emissions, rather than a proportional weakening of vertical transport.