Volcanic aerosol profiles at Lhasa were obtained during the 2019 Asian Summer Monsoon (ASM) season as part of the Sounding Water Vapor, Ozone, and Particles (SWOP) campaign. The SWOP campaign was led by the Institute of Atmospheric Physics, Chinese Academy of Sciences. Balloons were equipped with an electrochemical concentration cell (ECC) ozonesonde, a cryogenic frostpoint hygrometer (CFH), a compact optical backscatter aerosol detector (COBALD), and an International Met Systems (InterMet) iMet-1-RSB radiosonde . However, ECC ozone measurements are not analyzed in this study. A total of 9 measurements occurred in July–August (1 in July, 8 in August), followed by monthly measurements through January 2020. On 20 August 2019, the payload also included a Portable Optical Particle Spectrometer (POPS). During July–August 2019, Lhasa lay within the interior of the ASMA at UTLS levels (Fig. 1a). The site was chosen to observe the Raikoke plume in the ASMA interior and to document its arrival and vertical evolution. Additional information on SWOP field activities and on campaigns conducted in other years is available in .

The POPS data over Boulder were provided by the Baseline Balloon Stratospheric Aerosol Profiles (B2SAP) project; we analyze five profiles – 3 influenced by the Raikoke eruption (2 in August 2019 and 1 in November 2019) and 2 background references on 28 June 2019 and 3 December 2019 .

Figure 1a shows the geographic locations of Raikoke volcano and the balloon-sounding sites at Lhasa and Boulder. An overview of the analyzed balloon flights at Lhasa and Boulder, including the identified plume-layer ranges, is provided in Table 1.

The Compact Optical Backscatter Aerosol Detector (COBALD) was developed by ETH Zurich . Its lightweight and portable design makes it suitable for balloon payloads used in the study of cirrus clouds and aerosols . COBALD uses two high-power LEDs emitting at 455 nm (blue) and 940 nm (near‑infrared). It measures backscattered light from air molecules and particles, including aerosols and cloud particles (e.g. ice crystals). The backscatter ratio (BSR) is defined as 1BSR=βair+βparticlesβair, where βair is the backscatter coefficient of air molecules and βparticles is the backscatter coefficient of aerosols or cloud particles. The instrument measures only the total backscatter (βtotal=βair+βparticles); βair is calculated in post-processing from Rayleigh scattering using the measured pressure and temperature profiles, and βparticles is obtained as βtotal-βair. Here, BSR455 represents the backscatter ratio at 455 nm. The instrument's field of view is ± 6°, and its detection range spans 0.5–10 m. Because sunlight severely interferes with the measurements, COBALD is only operated at night. COBALD BSR uncertainty is typically characterized by an absolute error interval of about ∼ 5 % for the profile, while the precision is better than ∼ 1 % under UTLS conditions .

The COBALD color index (CI) is defined as 2CI=BSR940-1BSR455-1, which serves as an empirical proxy for particle size within aerosol layers. Note that when BSR455 and BSR940 are close to 1, CI can become sensitive to small baseline or calibration offsets.

The Cryogenic Frostpoint Hygrometer (CFH) is a high‐precision instrument for measuring water vapor based on the chilled mirror principle. It controls the mirror temperature to maintain a stable frost- or dew-point. Combined with temperature and pressure data from the iMet radiosonde, the relative humidity over ice (RHice) is calculated using the empirical equation from : 3RHice=eice(Tmirror)eice(Tenvironment), where eice(T) is the saturation vapour pressure over ice at temperature T (in Kelvin), Tmirror is the measured frost-point temperature, and Tenvironment is the ambient air temperature. CFH uncertainties may be as low as ∼ 2 % in the lower troposphere and ∼ 5 % near the tropical tropopause under good operating conditions . For UTLS conditions we use a conservative uncertainty of ± 10 % .

The Portable Optical Particle Spectrometer (POPS) is a lightweight instrument for measuring aerosol number density and size distribution . POPS employs a 405 nm laser to detect light scattered by individual particles, with scattering intensity related to particle size. It measures particles from 140 to 2500 nm and reports number concentrations in size bins . Measurement uncertainties are dominated by particle sizing (including sensitivity to the assumed refractive index) and flow-rate calibration .

TROPOMI, the satellite instrument aboard ESA's sun-synchronous Sentinel-5P platform launched in 2017, is a hyperspectral imaging spectrometer that records back-scattered radiation from the ultraviolet to the shortwave infrared . Total-column SO2 in Dobson units (1 DU = 2.69 × 10¹⁶ molec.cm-2) is retrieved using differential optical absorption spectroscopy (DOAS) applied to three wavelength windows (312–326, 325–335, 360–390 nm) . The publicly available Level 2 product provides four vertical columns: the surface–to–top-of-atmosphere column and three columns assuming an SO2 plume centered at 1, 7, or 15 km altitude. Following previous Raikoke studies , we adopt the 15 km retrieval, which most closely matches the eruption's mean injection height; values below the instrument's 0.3 DU detection limit are excluded, following . Four consecutive TROPOMI overpasses of the SO2 vertical column density at 15 km altitude captured the entire Raikoke SO2 cloud between 22:46 UTC on 24 June and 03:50 UTC on 25 June 2019 (Fig. 1b).

The Chemical Lagrangian Model of the Stratosphere (CLaMS) is a chemistry–transport model that calculates the three-dimensional motion of air parcels, which in turn represent the model grid . The CLaMS hybrid vertical coordinate (ζ) follows orography near the surface and transitions smoothly into θ once σ = p/psurf reaches 0.3 (usually about 300 hPa) . The isentropic coordinate in CLaMS is well suited for representing transport and mixing in the stratosphere. All CLaMS applications in this study – both backward trajectories and three-dimensional simulations with SO2-based tracers – are driven by ERA5 reanalysis: the standard runs use the native hourly fields on the 0.3° × 0.3° (latitude × longitude) grid with 137 vertical levels up to 80 km , whereas the sensitivity experiment in Appendix Fig. A2 repeats the simulation with ERA5 fields at 1° × 1° spatial and 6 h temporal resolution.

Diabatic backward trajectories are initialized every second along the balloon's vertical ascent profile, using the in-situ measurements of temperature (for deriving potential temperature), pressure, time, longitude, and latitude to define the start positions. Vertical velocities were computed from ERA5 wind fields and total diabatic heating rates, including clear-sky and cloud radiation, latent heat release, and turbulent and diffusive transport . This approach captures uplifts by resolved convection as represented in ERA5 . To identify the volcanic influence, only those backward trajectories that passed through the Raikoke eruption region (Fig. 1b) during 22:46 UTC on 24 June and 03:50 UTC on 25 June 2019 were retained.

Global three-dimensional transport simulations use passive SO2-based tracers initialized from the horizontal extent of the Raikoke SO2 cloud in the TROPOMI 15 km product (Fig. 1b; four overpasses between 22:46 UTC on 24 June and 03:50 UTC on 25 June 2019; values ≥ 0.3 DU) . At 00:00 UTC on 25 June 2019, parcels inside this mask are initialized to 1 (0 outside) within 3 potential-temperature ranges (380–400, 400–420, and 420–440 K) and then advected and mixed globally, yielding tracer fractions between 0 and 1.

To assess how different mixing intensities influence the reconstruction of volcanic plume transport processes, two simulations were conducted: (i) a control simulation with mixing every 24 h and λc=1.5 (a commonly used set-up for CLaMS simulations driven by high-resolution ERA5 reanalysis data; ); (ii) a modified simulation with mixing every 6 h and λc=3.5. In the CLaMS mixing scheme, parameterized mixing is triggered when the integral deformation between neighboring air parcels exceeds an empirical critical deformation γc=λcΔt, where λc is the critical Lyapunov exponent and Δt is the advective time step (mixing interval). For a given Δt, a larger λc (thus a larger γc) requires stronger deformation to trigger mixing and therefore corresponds to less frequent parameterized mixing (and vice versa) . In our setup, the modified simulation corresponds to enhanced parameterized mixing compared to the control simulation. Throughout most of the paper we show results from the modified simulation, as these agree better with the observations. Sensitivity to parameterized mixing intensity and comparisons with the control simulation are discussed in Sect. 5.1. The mixing configurations and the additional sensitivity runs (rectangular mask and coarser ERA5 input) are summarized in Table 2.

To verify that the enhanced aerosol layers observed over Lhasa are linked to the Raikoke eruption, we calculated ERA5-driven diabatic backward trajectories with CLaMS (Fig. 4). Trajectories were released every second along each balloon ascent within the selected plume-layer potential-temperature intervals (orange shading in Fig. 2), using the iMet-derived time and location (longitude, latitude, pressure, and temperature) as initial conditions. All trajectories were calculated backward to a single common reference time (21 June 2019, 18:00 UTC), which serves as the trajectory endpoint for all cases.

Pronounced BSR455 peaks were observed over Lhasa on 3 and 6 August 2019 (Fig. 2). For the upper layer on 3 August 2019, backward trajectories indicate that the air parcels were transported approximately along isentropic surfaces (Fig. 4). By contrast, the lower layer on 3 August 2019 includes trajectories lofted from ∼ 370 to ∼ 410 K. In the upper layer on 3 August 2019, the transport pathways differed significantly from other layers, as evidenced by changes in longitude (see Fig. 5 for trajectory details). This suggests that the enhanced BSR455 resulted from volcanic aerosols originating from two distinct branches of the volcanic plume at different altitudes, each following a different transport pathway. From 8–20 August 2019, mixing from lower potential-temperature levels began to occur. This coincided with a decrease in BSR455 values in the in-situ measurements, as air with relatively low aerosol concentrations dilutes the enhanced signals. Over the following three months, as the ASMA weakened seasonally, air from lower potential-temperature levels increasingly influenced the Lhasa profiles. During this period, peak BSR455 within the identified plume layer (orange shading in Fig. 2) decreased from values up to 1.88 in early August 2019 (95th percentile up to 1.77) to ≤ 1.25 from 30 September to 24 November 2019 (95th percentile ≤ 1.24), indicating progressive dilution of the volcanic aerosol layer by relatively aerosol-poor air from the lower troposphere.

To illustrate air masses influenced by the Raikoke eruption, we filtered backward trajectories to include only those passing through the eruption region (Fig. 1b) between 22:46 UTC on 24 June 2019 and 03:50 UTC on 25 June 2019. We then examined several representative pathways to show how the volcanic plume entered the ASMA interior (Fig. 5). In Fig. 5, the reported fraction is defined as (number of trajectories that meet the eruption-region criterion)/(total number of backward trajectories initialized for the corresponding flight date and θ layer shown in Fig. 4). For the 6 cases shown in Fig. 5, the total numbers of initialized trajectories are 79 and 50 (3 August 2019, two θ layers), 72 (6 August 2019), 190 (8 August 2019), 165 (10 August 2019), and 105 (12 August 2019). Because this filtering criterion is highly selective, only a small fraction of trajectories remains, and the resulting fractions can be regarded as conservative estimates. A sensitivity test shows that the fraction of backward trajectories reaching the eruption region is ∼ 3 %–10 % when using the original TROPOMI SO2 footprint and satellite-pass time window, but increases to ∼ 12 %–28 % when applying a broader rectangular domain (42–73° N, 137–215° E) and extending the time window to 7 d. The retained trajectories still follow the same main transport pathways shown here, providing confidence in the identified transport patterns.

For the calculation on 3 August 2019, the trajectories are divided into two distinct branches. The trajectories in the lower branch originate from 410–422 K and correspond to the main volcanic aerosol plume carried by the subtropical westerly jet. The upper branch originates from 459–472 K. Satellite observations from TROPOMI on Sentinel-5P indicate that the VVP core was entrained into the summertime easterlies around 20–25 July . The backward trajectories initialized in the 3 August 2019 plume layer (459–472 K) in Fig. 5 are consistent with the late-July phase of this satellite-tracked pathway. The potential temperature of the VVP during its transit through the ASMA inferred from satellite detections also closely matches the altitudes of enhanced BSR455 . Meanwhile, lidar measurements in Wuhan (central China) on 30 July reveal that the VVP's core backscatter coefficient is much greater than that of the main aerosol plume , implying that our measurements in this branch captured only the trailing filament of the VVP. Although both sets of air parcels belong to the same balloon profile, they trace back to two distinct plume components: the lower-θ main plume in the subtropical westerly jet (∼ 410–422 K) and the higher-θ VVP filament in the summertime easterlies (∼ 459–472 K). This indicates different source altitudes within the Raikoke plume and distinct dynamical transport pathways.

From 6 to 20 August 2019, the overall transport pattern remained similar to that on 6 August 2019 – corresponding to the main volcanic aerosol plume primarily driven by the subtropical westerly jet. After entering the ASMA, some air parcels deviate from the jet pathway and circulate within the anticyclonic region. This increased spreading of trajectories suggests dispersion and lateral mixing with surrounding air, which may contribute to the observed decrease in BSR455.

Backward trajectories for the Boulder flights on 7 and 27 August 2019 are shown in Fig. 6. On 7 August 2019, an enhanced aerosol layer is observed between 373 and 442 K (particle number concentration > 150 cm-3). The corresponding backward trajectories indicate transport to Boulder primarily along the subtropical westerly jet. On 27 August 2019, two volcanic layers are present. The upper layer (480–505 K) is associated with the VVP filament transported in summertime easterlies. The lower layer occurs at 385–429 K. To better illustrate the transport pathway, we plot only trajectories initialized between 405 and 410 K. These trajectories indicate that the plume had already traveled once around the globe before reaching Boulder.

In global three-dimensional CLaMS simulations, we use passive tracers initialized from the observed SO2 plume mask (Fig. 1b). We refer to these as SO2-based tracers. At 00:00 UTC on 25 June 2019, the tracer is set to 1 inside the mask (0 outside) within 3 potential-temperature ranges (380–400, 400–420, and 420–440 K) and then advected and mixed throughout the global atmosphere in CLaMS. We present results for the 400–420 K release, while the 380–400 and 420–440 K releases are used as sensitivity tests (Sect. 5.2; Figs. 11–12). Figure 7 shows global maps of fractions of the SO2-based tracer falling within the 400–420 K layer for each measurement day from 30 July to 20 August 2019, using the modified simulation. Here, the tracer fraction (0–1) represents the fractional contribution of air originating from the volcanic plume (identified by the TROPOMI SO2 mask shown in Fig. 1b) after advection and mixing.

The ASMA is indicated by the Montgomery streamfunction at the 378 × 10³ m2s-2 contour (layer mean over 400–420 K) as an estimate of its edge. The 11 PVU potential vorticity contour (layer mean over 400–420 K) illustrates the isentropic transport, including ASMA outflow (filaments) and in-mixing of PV-rich air from high latitudes. The strong transport barrier of the ASMA observed within ∼ 370–390 K is much weaker above 400 K . The results shown in Fig. 7 illustrate the overall intrusion of air masses into the ASMA and agree well with the Lhasa COBALD BSR455 observations. From 30 July to 1 August 2019, part of the SO2-based tracer entered the ASMA from its eastern side without reaching Lhasa. This is consistent with the absence of clear aerosol signals in the lower layers of the balloon-borne measurement data. On 3 and 6 August 2019, air masses impacted by the SO2-based tracer arrived above Lhasa. Without a significant effect of upwelling from below at that time, the SO2-based tracer signals remained sharp and concentrated. From 8 to 20 August, most of the ASMA contained mixed air masses. However, SO2-based tracer fractions inside the ASMA remained lower than outside, reflecting the anticyclonic transport barrier. During August, only a minor fraction of the SO2-based tracer is located inside the ASMA, while the majority remains outside. By later months, progressive mixing reduces the contrast between ASMA and surrounding regions in the tracer distribution (not shown here).

The corresponding SO2-based tracer fractions within the 400–420 K layer over the North American continent are shown in Fig. 8. Elevated tracer fractions are simulated over Boulder on both 7 and 27 August 2019. By 8 November 2019, the overall tracer fractions are markedly lower, reflecting progressive dilution by mixing during transport.

To assess the sensitivity of plume dispersion to model settings, such as parameterized mixing intensity, tracer release regions, and release altitudes, we performed a series of sensitivity experiments. In previous work with CLaMS driven by 1° × 1° resolution and 6 h temporal resolution reanalysis data (e.g., ERA5, ERA-Interim, as applied in ), the intensity of parameterized small-scale mixing was controlled by choosing a critical Lyapunov exponent λc=1.5 and a mixing interval of 24 h (e.g., ). Here, we performed simulations with CLaMS driven by high-resolution ERA5 data (0.3° × 0.3° resolution, 1 h) and found best agreement with observations by enhancing mixing intensity in the simulation slightly by choosing λc=3.5 and a mixing interval of 6 h. Taken together, the modified setup corresponds to enhanced parameterized mixing compared to the control setup.

To evaluate the sensitivity of our results with respect to parameterized mixing intensity, Figure A1 in the Appendix shows the SO2-based tracer fractions from the control simulation (λc=1.5; mixing interval Δt = 24 h) and compares them with those from the modified simulation (λc=3.5; Δt = 6 h) in Fig. 7. The control simulation in Figure A1 shows a similar overall intrusion pattern but lacks the distinct fractions over Lhasa on 3 August 2019. The tracer distribution in the control simulation is more fragmented, presenting small, isolated pockets compared to the more homogeneous structures in the modified simulation. In a further sensitivity test, we repeat the control simulation with ERA5 fields at coarser spatial and temporal resolution (details in Table 2). This setup reduces computing time and is often used for global CLaMS simulations that cover multiple years. However, changing the ERA5 resolution has only a minor effect and does not affect our conclusions. This is evident from the comparison of Fig. A1 (0.3° × 0.3°, 1 h) and Fig. A2 (1° × 1°, 6 h).

The different Raikoke tracers from the global three-dimensional CLaMS simulations are interpolated in time and space to the balloon-borne vertical profiles, providing a tracer-fraction profile at each measurement time. By comparing these tracer-fraction profiles from the model simulations with different mixing intensities with the observations (Fig. 9), we assess how the chosen parameterized mixing intensities affect the accuracy of the simulated plume dispersion. To quantify model–measurement agreement for each profile, we compute (i) the Pearson correlation coefficient r between the area-normalized tracer-fraction profile and the COBALD aerosol backscatter ratio at 455 nm (ABSR455), defined as BSR455-1 e.g.,, over the main plume layer (375–450 K; 375–475 K on 30 September, 28 October, and 24 November 2019) and (ii) the absolute peak-height difference |Δθ| between the two profiles. Note that r reflects only linear association and should be interpreted as a relative indicator rather than a definitive measure of agreement. The values of r and |Δθ| are annotated in Fig. 9. In Fig. 9, the SO2-based tracers are also released at 400–420 K. This setup primarily samples the diluted main plume rather than the higher-altitude trailing filament of the vorticized volcanic plume (VVP). Accordingly, the higher-altitude peak on 3 August 2019 likely originates from the VVP filament near 460–490 K and is not resolved by the 400–420 K release. A sensitivity simulation targeting the upper-level VVP filament shows enhanced tracer fractions near the observation region (not shown here); however, with the currently used CLaMS resolution, this thin and sharp aerosol peak at a single profile location is difficult to reproduce. Consistent with Figs. 7 and A1, the modified simulation produces tracer peaks (blue lines) that closely match the balloon measurements – except on 6 and 8 August 2019, when the peak altitudes deviate slightly. For most dates, the modified simulation shows consistently higher r and smaller |Δθ|. Although the control simulation (orange lines) occasionally aligns with observed peak altitudes, its overall agreement is weaker and it shows large peak-altitude biases on 12 August 2019 and 24 November 2019 (|Δθpeak| ≈ 25 and ≈ 36 K, respectively; Fig. 9). Overall, the modified simulation captures the vertical structure of the BSR455 profile well, despite minor differences in peak altitude for certain dates.

Furthermore, we assess the sensitivity of the model simulations to the plume initialization by carrying out another simulation using the modified configuration, in which Raikoke plume tracers are initialized within a rectangular latitude–longitude mask (from 49 to 62° N and 163° E to 170° W), following (see Table 2 for simulation details). Tracers were released on multiple isentropic levels within this rectangular domain. Here, we show the rectangle-based tracer fractions, released at potential‐temperature levels of 400–420 K (red lines in Fig. 9). The peak heights of the rectangle-based tracer fractions (red lines) and the SO2-based tracer fractions (blue lines) show no significant difference. Because the release domain captures the plume core, even a simple rectangular mask reproduces the observed peak altitude. Hence, for the Raikoke simulations the choice of injection region is less critical than the choice of model mixing intensity.

Over Boulder, we likewise compare the tracer-fraction profiles on the measurement dates with the POPS aerosol number concentration profiles at STP (Fig. 10). For the computation of r, we evaluate within 375–450 K for 7 and 27 August 2019, and within 375–475 K for 8 November 2019. Overall, the SO2-based tracers in the modified simulation (blue) reproduce the observed peak structure more closely, with higher r and smaller |Δθ| than the control and rectangle-based simulations.

The Raikoke plume was injected primarily within 8–16 km . The main injection occurred between 18:00 UTC on 21 June and 03:00 UTC on 22 June 2019 (with SO2 release possibly extending to ∼ 06:00 UTC) . Radiative heating of ash and nascent sulfate then lofted parts of the plume, raising volcanic cloud tops above 20 km within 4 d of the eruption . These observations indicate that Raikoke's plume experienced an intense updraft during the eruption stage. Aerosol‑radiative lofting is not included in the CLaMS simulations. However, comparison of the vertical profiles of the SO2-based tracers with in-situ measurements allows estimation of the lofting height. SO2-based tracers were injected at 20 K intervals between 360 and 500 K to cover the entire possible extent of the Raikoke plume. To assess which initialization range best reproduces the observed plume, we compare tracer‑fraction profiles with the BSR455 profiles in Fig. 11.

Overall, tracers initialized within 400–420 K (blue lines; ∼ 15–16.5 km at the Raikoke site) best match the enhanced BSR455 profiles, compared with 380–400 K (brown) and 420–440 K (purple). For most dates, the 400–420 K initialization attains the highest r and the smallest |Δθ| among the three layers (see annotations in Fig. 11). For instance, on 3, 6, and 12 August 2019, tracers initialized in the 400–420 K layer contribute most strongly to the peak of the BSR455 profile, while those from the 380–400 and 420–440 K layers also produce peaks at the same altitude but with smaller amplitudes. On 8 August, 30 September, and 28 October 2019, only the 400–420 K tracer reproduces the BSR455 peak at the correct altitude, whereas the peaks from the 380–400 and 420–440 K tracers appear below or above the observed height. For Boulder (Fig. 12), the 400–420 K initialization likewise shows the best agreement with the observations. On 7 August 2019 the 380–400 K layer, and on 27 August 2019 the 420–440 K layer, produce peaks that disagree strongly with the observations (r<0.1).

estimate a peak injection altitude near 11 km, whereas the tracer level that best matches the observed profiles corresponds to ∼ 15–16.5 km (400–420 K). We hypothesize that this ∼ 4–5 km offset could be consistent with radiative self-heating lofting of the SO2-rich cloud, but it may also reflect uncertainties in injection-height estimates across techniques and model-related discrepancies (e.g., ).