Conducted from June through September 2022 in the Houston, Texas, region, the TRACER Intensive Observation Period (IOP) was timed to capture frequent warm-season convection generated by strong surface heating, abundant Gulf moisture, and sea-breeze circulations. Combined with Houston's urban and industrial emissions and the contrasting marine and rural air masses, these conditions provided an ideal setting for investigating aerosol–convective-cloud interactions across a wide range of aerosol loadings and meteorological conditions (Jensen et al., 2025). During the IOP months, the ARM TBS was deployed for the first 2 weeks of each month at an ancillary (ANC) site in Guy, Texas (29.33° N, 95.74° W), southwest of Houston. Concurrently with TBS flights, a co-located 3-channel RPG-G5 microwave radiometer provides time-series measurements of brightness temperature from three channels centered at 23.8, 30, and 89 GHz (Cadeddu et al., 2013), which are used to retrieve precipitable water vapor (PWV) and liquid water path (LWP) (Turner et al., 2007). This location was selected to represent a rural background with less influence from a mix of aerosol sources (such as fresh aerosol emissions from the metropolitan Houston area).

Complementing the stationary measurements conducted by the ARM TBS team, the collaborating Texas A&M University (TAMU) team undertook mobile sampling of the sea breeze. Their sampling locations, illustrated in Fig. 1, typically originated on the coast in Galveston. From there, they followed the sea breeze inland, concluding at either the AMF site in La Porte or at one of several sites west and northwest of Houston (Bear Creek, Hockley, or Hempstead) (Thompson et al., 2025a). The specific transect routes are not shown because they varied among sampling days. Along this mapped corridor, TAMU mobile measurements frequently transitioned from marine boundary-layer air characterized by low particle concentration to continental air ahead of the sea-breeze front, which generally contained particles that were slightly more effective for cloud formation and exhibited substantial variability over tens of kilometers (Thompson et al., 2025a, b; Chen et al., 2025).

Additionally, the Next Generation Weather Radar (NEXRAD), designated KHGX and located at 29.47° N, 95.08° W, played a pivotal role in observational activities throughout the study period (Galfione et al., 2025). This advanced radar system conducted volume scans using Plan Position Indicator (PPI) techniques, systematically operating over elevation angles from 0.5 to 19.5°. By varying these elevation angles, the radar prioritized acquiring data essential for detecting and characterizing convective processes embedded within stratiform regions. This approach ensured comprehensive coverage and improved the accuracy of convective feature identification, which is critical for understanding aerosol–cloud interactions and precipitation patterns in the Houston area.

The TBS observations enable examination of vertical variations in aerosol properties, including changes in concentration and composition driven by sources such as urban pollution, industrial emissions, long-range transport, and marine influences. The ARM TBS (Dexheimer et al., 2026) uses winch-controlled, helium-filled aerostats to obtain in situ vertical profiles up to about 1.5 km above ground, operating below clouds. Typical payloads, as shown in Table 1 (a payload picture in Fig. S1 in the Supplement), include sensors for pressure, temperature, and humidity (e.g., iMet-XQ2), an anemometer for wind speed and direction, and a portable optical particle spectrometer (POPS) and a condensation particle counter for aerosol size distributions and concentrations; a Size- and Time-resolved Aerosol Collector (STAC) collected the aerosol particles at various altitudes for offline chemical analysis. Both the POPS and CPC were equipped with dryers in their sampling lines to reduce the relative humidity of the sampled aerosol flow to below 40 %. The POPS size bins were calibrated using DMA-classified ammonium sulfate particles rather than polystyrene latex spheres, so the reference refractive index of the sizing scale (n ≈ 1.54 at λ = 405 nm) falls within the range reported for ambient submicron aerosol, reducing the sizing bias associated with the assumed refractive index. A further consideration is the influence of long-range-transported mineral dust, whose refractive index differs from that of the ammonium sulfate used to assign the POPS size bins. We therefore assessed the sensitivity of the retrieved size distribution to the assumed refractive index. Adopting a dust-representative value (m = 1.52 + 0.002i) in place of the PSL calibration value and recomputing the Mie response curve shifts the bin boundaries by less than 4 % in diameter below 1.38 µm (Kezoudi et al., 2026), These shifts are smaller than the corresponding POPS bin widths across the size range; i.e., within the instrument's size resolution, the shape of the binned distribution is preserved to within counting uncertainty. This indicates that the change in CCN concentration is small.(Mei et al., 2024, 2022) During the TRACER IOP, a subset of meteorological and aerosol measurements was collected from 149 ARM TBS flights, which were conducted in the first 2 weeks of each month from June to September. The monthly distribution included 46 flights on 11 d between 3 and 14 June (46.45 h), 32 flights on 13 d between 2 and 14 July (41.12 h), 43 flights on 13 d between 2 and 14 August (46.92 h), and 28 flights on 12 d between 2 and 14 September (56.2 h). Detailed TBS daily operation reports can be found in Appendix A of the campaign report (Jensen et al., 2023). The taking-off and landing times for each flight were listed in Table S1 in the Supplement.

During the 2022 deployment, aerosol particles were collected using a 4-stage impactor (Sioutas) at ground level and a STAC (Cheng et al., 2022) on a TBS. The instrument collected size- and time-resolved samples directly onto substrates for two complementary analytical pathways at the Environmental Molecular Sciences Laboratory (EMSL, https://www.emsl.pnnl.gov/, last access: 22 November 2025). For single-particle characterization, particles collected at stage D (50 % cutoff size of 0.12 µm) of the STAC sampler were analyzed using Computer-Controlled Scanning Electron Microscopy with Energy-Dispersive X-ray spectroscopy (CCSEM-EDX). This technique yields individual particle size, morphology, mixing state, and elemental composition, helping identify particle types and sources (Lata et al., 2023). For molecular-level characterization, samples were collected on Teflon filters for about 12 h at ground level and analyzed by nanospray desorption electrospray ionization high-resolution mass spectrometry (nano-DESI-HRMS) (Gautam et al., 2024). The resulting mass spectra inform the constituent organic molecular formulas in the particle phase, from which we derive key parameters such as O:C ratios, carbon oxidation state, aromaticity indices, and organic aerosol volatility (Roach et al., 2010; Vandergrift et al., 2024; Vandergrift et al., 2022). The results discussed herein are from samples collected during the June and September 2022 flights.

At the ANC site, additional ground-based instruments were deployed to support the measurements conducted at the AMF site. The deployed instruments and the corresponding measurements are included in Table 2.

A range of ground-based measurements was obtained from the ARM Mobile Facility (AMF) site at La Porte, TX. The AMF was equipped with advanced radar and lidar systems, aerosol-observing instruments, and meteorological sensors. ARM has developed a set of Planetary Boundary Layer Height (PBLHT) value-added products to help scientists evaluate and compare estimates. In this study, the Planetary Boundary Layer Height Radiosonde Retrievals with yearly output (PBLHTSONDEYR1MCFARL) VAP was used to provide the PBLHT (Sivaraman et al., 2013; Li et al., 2021; Liu and Liang, 2010).

Additionally, this study used aerosol vertical profiles retrieved from a micropulse lidar (MPL) using a newly developed technique by the Brooks group at Texas A&M University (TAMU) that combines ground-based aerosol measurements with lidar and radiosonde data to obtain vertical profiles of aerosol, CCN, and INP (Chen et al., 2025). Chen's retrieval method assumes that surface measurements of aerosol size distribution, composition, and cloud-activating ability are representative of their vertical profiles, except in cases of elevated layers like smoke or dust. The approach involves determining the vertical profile of the cloud-free aerosol backscatter coefficient using lidar and radiosonde data. The aerosol backscatter coefficient is then corrected for hygroscopic growth using the lidar hygroscopic growth correction factor f(RH), which is which is calculated from hygroscopicity parameters derived from ground-based Scanning Mobility Particle Sizer (SMPS), POPS, and CCN counter measurements, along with relative humidity profiles from radiosondes. The resulting dry aerosol backscatter profiles were subsequently used to retrieve vertical CCN concentration profiles. We used these retrieved profiles to complement our CCN profiles estimated from the ANC location and to provide broader spatial and vertical context.

To estimate CCN concentration, our method combines the observed TBS aerosol size distribution with ground-based chemical composition measurements. Our approach includes two steps. The first step is to address the size-range limitation of the aerosol size-distribution data. We extrapolate the measured particle number size distribution from its original POPS range of 135–3000 to 10 nm. We employed a modal-fitting technique based on the algorithm detailed by Seinfeld and Pandis (2016), Hussein et al. (2005). The complete size distribution was represented as the sum of up to two lognormal modes, each defined by its total number concentration (Ni), geometric mean diameter (Dgi), and geometric standard deviation (σgi). The parameters of the fitted size distribution were determined through iterative optimization, minimizing the sum of squared errors between the fitted and measured size distributions over the 135–1000 nm range. The fit was further constrained such that the integral of the distribution from 10 to 135 nm reproduced the difference between the total number concentrations measured by the CPC and the POPS to within 10 %, consistent with the combined counting uncertainty of the two instruments. Consequently, the integral of the fitted distribution over the full 10–1000 nm range agrees, within the same tolerance, with the total number concentration measured independently by the CPC.

The second step is to determine the CCN concentration profile using the κ-Köhler theory. In this study, we assume that the aerosol chemical composition at the ground was representative of that at each TBS flight altitude when the same airmass influenced the TBS flight. Aerosol composition is not necessarily vertically uniform, even within a nominally well-mixed boundary layer. New particle formation has been observed to occur preferentially in the upper mixed layer, the entrainment zone, and the residual layer rather than at the surface, where condensation sinks are larger, and precursor photochemistry differs. Semi-volatile species repartition in response to the vertical temperature and relative humidity gradients: ammonium nitrate and semi-volatile organics shift toward the particle phase at the lower temperatures encountered aloft, so the organic-to-inorganic ratio measured at the surface may not apply at the top of the profile (Zhang et al., 2026; Lata et al., 2025, 2023; Creamean et al., 2021; Huffman et al., 2009). Our assumption is therefore most robust for the daytime and convectively well-mixed boundary layer, and more uncertain for decoupled profiles, such as cases on 6 September (discussed more in Sect. 3.5). Ground-level non-refractory PM1 was measured with a Time-of-Flight Aerosol Chemical Speciation Monitor (ToF-ACSM, Aerodyne Research) equipped with a standard vaporizer (Li et al., 2025; Fröhlich et al., 2013), using a PM2.5 cyclone upstream of the inlet. The instrument provided 10 min-resolution mass concentrations of organics (Org), sulfate (SO4), nitrate (NO3), ammonium (NH4), and chloride (Chl) to complement the TBS deployment. However, the ACSM only quantifies non-refractory submicron species and cannot detect refractory components such as black carbon, sea salt, and mineral dust. These missing species, together with the assumption of a vertically uniform chemical composition, can influence the bulk hygroscopicity and introduce additional uncertainty into CCN concentration estimates – a source of uncertainty that warrants further experimental study.(Schulze et al., 2020; Ren et al., 2026)

We averaged the volumetric fractions of aerosol chemical composition over the flight period. Aerosol particles were assumed as internal mixtures, and the particle hygroscopicity (κCCN) is therefore the volumetric average of the three participating species. More information can be found in our previous publication (Mei et al., 2024) 1κCCN=∑ixiκi=xorgκorg+x(NH4)2SO4κ(NH4)2SO4+xNH4NO3κNH4NO3 where xi is the volume fraction of species i. The values of κ are 0.15, 0.67 and 0.61 for organic, (NH4)2SO4 and NH4NO3, respectively. The densities of (NH4)2SO4 and NH4NO3 were assumed to be 1770 and 1730 kgm-3, respectively. Organics were assumed to have a density of 1350 kgm-3, which is within the typical range of previous measurements.(Engelhart et al., 2011)

For a given supersaturation (SS), the κ-Köhler theory is used to determine a corresponding critical activation diameter Dpc, 2Dpc3=4A327κCCN(SS)2 where A=4σwMwRTρwD, Mw the molecular weight of water, σw the surface tension of pure water, ρw the density of water, R the gas constant, and T the absolute temperature.

All particles in the measured size distribution larger than this critical diameter are assumed to activate into cloud droplets. The total CCN concentration is then found by integrating the particle size distribution from this critical diameter upwards.

The ARMTRAJ dataset (Silber et al., 2025b) is generated using the Hybrid Single-Particle Lagrangian Integrated Trajectory (HYSPLIT) model informed by the fifth-generation European Center for Medium-Range Weather Forecasts (ECMWF) atmospheric reanalysis dataset (ERA5) (Hersbach et al., 2020; Stein et al., 2015). The ERA5 dataset is utilized at its highest spatial resolution (0.25°; ∼ 31 km) and 1 h temporal resolution. The core value of ARMTRAJ is to bridge the gap between the fixed, or Eulerian, nature of ARM's ground-based measurements and the need to understand air-mass hysteresis as it passes over the site. To quantify uncertainty, the dataset incorporates results from ensemble runs for each trajectory initialization; the variability within these ensembles serves as a valuable metric for assessing the consistency and robustness of the reported air-mass coordinates and state variables (Silber et al., 2025b, a). For the ARMTRAJ-TBS, the trajectories were estimated at 11 equally spaced altitudes between the lowest and highest altitudes probed by the balloon, over 1 h intervals (Silber et al., 2025a).

To identify the dominant patterns of meteorological variability within clustered atmospheric regimes indicated by the trajectory dataset, we performed Principal Component Analysis (PCA) on it. Principal Component Analysis (PCA) was employed to reduce the dimensionality of the dataset while retaining as much of the variance as possible from the original variables. PCA is a multivariate statistical technique that transforms a set of possibly correlated variables into a smaller number of uncorrelated variables known as principal components (Jolliffe and Cadima, 2016). These components are linear combinations of the original variables and are ordered so that the first principal component accounts for the largest possible variance in the data, with each successive component capturing the maximum remaining variance subject to the constraint of being orthogonal to the preceding components. Prior to analysis, the data were standardized to ensure that contributions were not biased by scale, thereby allowing the PCA to reveal more meaningful underlying structures. In this study, the first two principal components (PC1 and PC2) accounted for more than 80 % of the total variance in each regime. The weights for these components reveal distinct physical processes.

Using air-mass geospatial trajectory data derived from the TBS latitude and longitude coordinates via the ARMTRAJ dataset, Fig. 2 illustrates clustered atmospheric transport pathways over two time periods: 24 h (Fig. 2a) and 5 d (Fig. 2b). K-means clustering was applied to classify trajectories into three groups, highlighting contrasting transport pathways and source regions. For the 24 h back trajectories, Cluster 1 (blue) comprises air masses that travel almost exclusively over the Gulf of Mexico before reaching the ANC site, indicating a relatively direct marine inflow. In contrast, Cluster 2 (orange) follows more complex marine pathways: the trajectories initially traverse the Gulf, then curve inland and pass over land for part of their transport, indicating a mixed marine–continental influence and a greater potential for modification by boundary-layer processes over land. Cluster 3 (yellow) originates primarily from the north and west, reflecting stronger influence from continental sources, including anthropogenic emissions from the broader Houston metropolitan region and upwind areas. The 5 d analysis (Fig. 2b) provides crucial insight into the long-range history of these air masses. It confirms that Clusters 1 and 2 spent the vast majority of their transit time over the ocean, with pathways extending deep into the Gulf of Mexico and the western Atlantic. Conversely, the 5 d trajectories for Cluster 3 reveal a significant, long-range continental influence, with air masses originating from the central and upper Midwest of the United States and Canada. This demonstrates that while Clusters 1 and 2 are primarily influenced by oceanic processes, Cluster 3 is shaped by prolonged transport over land, bringing distinctly continental air to the site.

Further analysis using PCA provides additional insights into the meteorological conditions for each cluster. As shown in Fig. 3, the component weights illustrate the influence of eight meteorological variables on PC1 and PC2. The color indicates whether a variable's contribution is positive (yellow) or negative (red). The variables analyzed include airmass pressure (pres), potential temperature (theta), temperature (temp), planetary boundary layer height (pblh), relative humidity (rh), specific humidity (qv), equivalent potential temperature (theta_e), and the mean hourly airmass ascent rate (wvert).

In the marine-influenced regime (Cluster 1), the PCA cleanly separates the advective characteristics of the marine boundary layer from local diurnal forcing. The leading component (PC1) is strongly associated with planetary boundary layer height (pblh, -0.45), potential temperature (theta, -0.34), relative humidity (rh, 0.48), and specific humidity (qv, 0.46). We therefore interpret PC1 as an index of marine layer strength and intrusion. Positive PC1 scores correspond to a cool, moist, and shallow marine boundary layer characteristic of direct onshore flow, whereas negative scores indicate a deeper, warmer, and drier boundary layer associated with marine boundary layer erosion or increased continental influence. The second component (PC2) is dominated by positive weights for pblh (0.63), temperature (temp, 0.46), and equivalent potential temperature (theta_e, 0.41). Because PC2 is orthogonal to the marine-layer signal, it is interpreted as representing the diurnal cycle of surface heating and boundary-layer growth.

For the regime influenced by anthropogenic emissions and transport (Cluster 3), the PCA identifies atmospheric dispersion capacity as the primary mode of variability. The leading component (PC1) is dominated by pblh (0.71) and is strongly correlated with rh (-0.52). In this regime, PC1 functions as a ventilation index, quantifying the atmosphere's capacity to dilute surface emissions. High positive scores correspond to a deep, well-mixed boundary layer and reduced near-surface humidity, consistent with vigorous vertical mixing. Negative scores indicate a shallow, stable boundary layer that traps pollutants and moisture near the surface, favoring episodes of poor air quality. The second component (PC2) is characterized by positive weights for surface pressure (pres, 0.57), temperature (temp, 0.47), and specific humidity (qv, 0.47), and likely corresponds to conditions within a warm, moist high-pressure system.

Under mixed-influence conditions (Cluster 2), the leading mode of variability (PC1) is overwhelmingly characterized by strong, co-varying positive weights for specific humidity (qv, 0.58) and equivalent potential temperature (theta_e, 0.58). This component serves as a robust proxy for the large-scale advection of moisture and thermodynamic energy into the region. The second component is almost exclusively defined by a large weight for pblh (0.84) and a strong correlation with rh (-0.42). This signature suggests vertical mixing and entrainment, in which a growing boundary layer mixes with drier air from the free troposphere, thereby increasing pblh while reducing the layer-averaged rh.

In summary, the two-component PCA, stratified by regime, isolates the dominant atmospheric controls: marine-layer confinement versus diurnal heating (Cluster 1), moisture/energy advection versus entrainment-driven drying (Cluster 2), and boundary-layer ventilation capacity versus synoptic airmass forcing (Cluster 3).

To contextualize the summertime boundary-layer behavior, we characterized its diurnal evolution using radiosonde profiles launched from the AMF site and two established algorithms implemented in the ARM PBLHT VAP product (Sivaraman et al., 2013). Estimating the Planetary Boundary Layer Height (PBLHT) from sounding data is a classic approach to understanding the PBL structure. However, no single perfect method exists, and the optimal approach varies with the definition; therefore, a given approach may be a better fit depending on the goal. Data from the ARM program enable comparison of several radiosonde-based approaches, including the Liu-Liang and bulk Richardson number methods (Li et al., 2021; Stull, 2012; Seidel et al., 2010; Liu and Liang, 2010), which are particularly effective for studying the diurnal cycle. The bulk Richardson number method, which identifies the PBL top as the point at which turbulence is dynamically suppressed, is physically robust and well-suited to determining the height of the nocturnal Stable Boundary Layer (SBL) and the transition periods. In contrast, the Liu-Liang method defines the PBLHT as the height at which the maximum vertical gradient in thermodynamic variables occurs, a feature characteristic of the entrainment zone atop a convective boundary layer.

As shown in Fig. 4, the mean diurnal evolution of the PBLHT during the summer and early autumn months (June–September) aligns with theoretical expectations – increasing over the course of daylight hours coupled with the diurnal cycle of surface temperature. The PBLHT determined by the Heffter method is systematically higher than that obtained by other methods. Because it identifies the PBLHT by locating the base of the elevated temperature inversion and using empirical thresholds that require tuning across regions, it might not be robust in this case for moist/marine regimes. A comparison between the Liu-Liang and Richardson number methods (with a critical value of 0.25) reveals strong agreement during daytime hours (05:30 (00:30), 11:30 (06:30), 17:30 (12:30), and 23:30 (18:30) UTC (local time). The convergent estimates during this period indicate that both methods effectively capture the same fundamental process: the growth of a convective mixed layer driven by surface heating. The Richardson method illustrates a pronounced collapse of the PBLHT, while the Liu-Liang method shows a much more gradual decrease. This divergence is physically instructive. The Richardson method, being a direct measure of turbulence stability, captures the rapid decay of buoyancy-driven mixing as solar heating ceases, signaling the swift transition to a shallower, shear-driven stable layer. The Liu-Liang method, by seeking the maximum gradient, is prone to identifying the top of the now-inactive, neutrally stratified residual layer left over from the afternoon, thus overestimating the height of the active nocturnal boundary layer.

Although the sounding data from the middle of the night were limited to fully characterize the developed SBL, the 05:30 CDT data point is particularly instructive for inferring the preceding nocturnal conditions. The trend of a progressively shallower early-morning PBL from summer to autumn is evident. The mean PBLHT at 05:30 decreased from approximately 500–600 m in June and July to 300–400 m in August, and further to around 200–300 m in September. This implies that the nocturnal SBL becomes systematically shallower and more stable as the season progresses from summer to autumn, a trend consistent with longer nights and more extensive radiative cooling, which fosters the development of stronger, more vertically constrained surface inversions.

To assess how well the TBS profiled the summertime boundary layer, we overlaid flight altitudes with the mean PBLHT derived from the Liu–Liang algorithm. The overlay (Fig. S2 in the Supplement) shows that most daytime flights in June–August remained below the mean Liu–Liang PBLHT, whereas several September flights reached into and, at times, above the mean PBL top, as shown in Fig. 5. This is primarily explained by the systematic seasonal decrease in PBL depth from mid-summer to early autumn. Operational factors also contributed: earlier in the summer, the requirement to operate below the cloud base often constrained the TBS to lower altitudes, whereas in September, clearer conditions allowed the balloon to ascend to its maximum altitude. As established in the previous analysis, the mean afternoon PBLHT is significantly deeper in June and July (often exceeding 1.5–2.0 km) compared to September, when it typically peaks around 1.0–1.3 km. This is a direct consequence of reduced solar insolation and shorter day lengths, which lessen the surface heating that drives convective turbulence and PBL growth. With a fixed or operationally limited TBS ceiling, flights were therefore sampled primarily within the well-mixed layer in JJA. Within this regime, temperature, humidity, and aerosol properties are vertically homogeneous to first order, and profiles are representative of surface conditions and daytime dilution. The data collected in JJA are thus well suited to characterizing mixed-layer chemistry and aerosol processing and to relating CCN to surface emissions.

To link air-mass origin with boundary-layer dynamics, we combined back-trajectory cluster analysis with radiosonde-derived PBL heights (Liu–Liang and Ri=0.25) and TBS flight altitudes to diagnose how source regions are introduced into, and mixed within, the summertime boundary layer. As the convective boundary layer grows during the day (often exceeding 1–1.5 km in JJA), entrainment at the PBL top brings the Cluster-2 air downward, continuedly diluting the near-surface air residue. Because most TBS flights in JJA remained below the mean PBLHT, they primarily sampled the well-mixed local layer. In September, the air mass above 500 m becomes more heterogeneous, with both Cluster 2 and Cluster 3 appearing aloft, gradually mixed with the near-surface Cluster 1 layer. The decline in PBLHT (shallower afternoon maxima and lower morning depths) lowers the entrainment zone. September flights often reached and exceeded the mean PBLHT, enabling direct sampling of the entrainment zone and the lower free troposphere, where the influences of Clusters 2 and 3 are strongest. With a shallower morning SBL (∼ 200–300 m) and reduced daytime growth in September, the boundary layer is more vulnerable to changes in synoptic and mesoscale flow (e.g., shifting trajectories, sea-breeze passages, residual-layer interactions). Consequently, all three clusters can influence the air below 500 m, with the dominant cluster varying from day to day. This variability might be caused by intermittent entrainment of the overlying Cluster 2 layers into a shallow mixed layer and the occasional persistence of the previous day's residual layer aloft.

Air mass transport pathways, origins, and processes profoundly affect aerosol distribution, particularly vertical variability. To understand how different atmospheric regimes influence the vertical aerosol structure, we grouped the flight data into three distinct clusters, as discussed in Sect. 3.1. Cluster 1 affected 27 d, Cluster 2 influenced 12 d, and Cluster 3 impacted 9 d. Note that 6 September flights were influenced by both Cluster 1 and Cluster 2 and were excluded from the cluster-influenced figures. Figure 6 shows the mean vertical profiles of temperature (T), relative humidity (RH), and aerosol number concentrations (Npops > 135 nm and Ncpc > 10 nm) composited by three trajectory-based influence clusters: Cluster 1 (marine-influenced), Cluster 3 (anthropogenic/transport influenced), and Cluster 2 (mixed influence). The aerosol number concentrations were reported at the standard temperature and pressure (STP; 1013.15 hPa and 298.15 K). In Fig. 6, plots represent daytime TBS profiles during the intensive period in 2022, and error bars denote one standard deviation about the mean in each 75-m altitude bin for each cluster. The vertical profiles of mean diameters and geometric standard deviations of the modes of merged CPC-POPS size distributions are included in Fig. S4 in the Supplement.

Subtle differences in the vertical profiles of temperature and relative humidity were observed among the three air-mass clusters. However, these differences may also reflect local cloud cover, solar heating, wind, soil moisture, and boundary-layer mixing effects. Thus, the small, discernable difference does not establish that air-mass origin alone caused it. The temperature profiles are broadly similar across all three clusters, with comparable lapse rates from the surface to approximately 1.5 km. This suggests that the local-scale thermal structure and boundary-layer depth were not strongly dependent on air-mass origin during the sampled period, as confirmed with the normalized vertical profiles in Fig. S3a in the Supplement. However, there are slight differences near the surface: the anthropogenic cluster (Cluster 3) is slightly cooler (∼ 30°C) than the marine (Cluster 1, ∼ 32°C) and mixed (Cluster 2, ∼ 31°C) clusters. The relative humidity (RH) profiles show similar variation. Cluster 3 is slightly more humid throughout the lower atmosphere, with RH values sustained at ∼ 60 % below 1 km, consistent with a moist, coastal air mass. Notably, the RH in Cluster 2 shows a weak positive gradient: starting from around 60 %–65 % near the surface, RH gradually rises with height, reaching its highest values (≈ 70 %–75 %) near 1.3–1.4 km. In Cluster 3, RH aloft behaves differently. It is comparable to the other two cluster-influenced conditions within the lowest 1 km, then declines above ∼ 1.2 km. This implies that the anthropogenic cluster is characterized by a moist, well-mixed lower boundary layer capped by relatively drier air aloft, which is typical of a polluted, coastal boundary layer capped by subsiding or advected drier free-tropospheric air.

The vertical profiles of aerosol concentrations and size distributions reveal clear distinctions between the three air mass clusters. Cluster 1 (marine) exhibits low aerosol concentrations characteristic of clean background air. Both Ncpc (> 10 nm) and Npops (> 135 nm) profiles show a uniform concentration up to ∼ 1.1 km. This vertical structure is consistent with a well-mixed marine boundary layer, in which the dominant aerosol sources and precursors originate from the ocean surface (e.g., sea spray and volatile dimethyl sulfide). Furthermore, the gradual decrease with height at altitudes above 1000 m is likely due to vertical mixing and/or dilution from entrainment of even cleaner free-tropospheric air from above. Cluster 3 shows a similar uniform concentration in Ncpc (> 10 nm) and a gradual decrease in the aerosol concentration of Npops (> 135 nm) as altitude increases from 100 m to 1.5 km. This indicates that, in both the clean marine (Cluster 1) and polluted anthropogenic (Cluster 3) cases, the dominant aerosol sources are located within the lower boundary layer. However, Cluster 3 (anthropogenic), influenced by human activities such as industrial emissions, is also characterized by higher concentrations of larger particles (Npops (> 135 nm)), while the total aerosol number concentrations (Ncpc (> 10 nm)) are comparable to those in Cluster 1. This combination suggests an aged, processed aerosol source. As shown in Fig. S4, the contrast in Naccum between Cluster 1 and Cluster 3 is by a factor of 4–5 while Dg,accum differences of only ∼ 30 % indicate that the marine and continental air masses over the ANC site differ mainly in the number concentrations of accumulation-mode particles, but not in the mean diameters. This observation suggests that the aerosol particles influenced by Cluster 1 are coastal-modified marine aerosol, but not pristine.

Cluster 2 (mixed) presents a complex vertical structure. At lower altitudes, the aerosol concentration profiles are relatively constant below 800 m. However, above 1 km, the Ncpc (> 10 nm) and Npops (> 135 nm) profiles show a notable increase, making it the most concentrated cluster aloft. This elevated enhancement is usually consistent with an advected, or long-range transported aerosol layer residing near or just above the boundary-layer top. Such layers often arise from regional or continental pollution, biomass burning, or other distant sources and can be decoupled from the local surface emission regime. The fact that Cluster 2 is intermediate in concentration near the surface but enhanced aloft supports the interpretation of a mixed air mass influenced simultaneously by local fresh emissions, and an overlying transported pollution layer entrained into the boundary layer. In addition, the Cluster 2 result is the strongest corroboration of the trajectory analysis. Aqueous-phase sulfate production in non-precipitating cloud is the most efficient mechanism for shifting the accumulation mode to larger sizes, and Fig. S4 shows the largest Dg,accum while carrying only an intermediate number is precisely the signature of a processed rather than a freshly emitted population.

Figure 7 presents vertical profiles of estimated CCN concentrations for three distinct air mass clusters, plotted as a function of altitude. These plots provide useful vertical context for the variation in CCN availability across airmass regimes. Another version of this figure was plotted against altitude normalized by the PBLHT (Altitude/PBLHT), as shown in Fig. S5 in the Supplement. This normalized version aims to reveal key differences in the vertical distribution of aerosols relative to the PBL structure across differing airmass influences. In Figs. 7a–c and S5, the size of the circle symbols at each altitude level is proportional to the number of data points collected within that altitude range, providing a visual indication of the statistical robustness of measurements. A threshold of 2000 data points is established as the confidence limit for data representativeness. Altitude bins containing fewer than 2000 data points (represented by smaller circles) should be interpreted with caution, as they may be less representative of the true vertical CCN distribution. Figures 7d and S3d appropriately document the sampling limitations inherent in TBS operations. They show a strong reduction in the number of samples collected aloft, and the data density decreases substantially above ∼ 800–1000 m altitude in Fig. 7d. In Fig. S5d, Cluster 2 shows notably fewer samples at the high normalized ratio (Altitude/PBLHT > 0.8) compared to Clusters 1 and 3. Because statistical confidence in the upper-level observations is correspondingly reduced, the figures are most suitable for discussing within-PBL gradients and airmass differences but provide less definitive information about the free-tropospheric tail.

The vertical structure in all three cluster cases generally shows a well-mixed boundary layer (< 800 m), characterized by relatively uniform CCN concentrations. Cluster 1 (Marine-influenced) is characterized by the lowest overall CCN concentrations, consistent with a relatively clean marine air mass originating from the Gulf of Mexico. Within the lower PBL (Altitude/PBLHT ≲ 0.4), mean CCN concentrations at SS = 0.80 % are on the order of 1000 cm-3. The ratio of CCN at 0.80 % to 0.13 % supersaturation is around 3 in this region, indicating relatively homogeneous critical supersaturation among aerosol particles, likely dominated by sea salt and marine biogenic particles with similar hygroscopic properties. The low ratio also suggests fewer small Aitken-mode particles relative to the accumulation mode, which is typical of clean marine environments where new particle formation is suppressed or where coagulation/wet scavenging removes smaller particles.

Cluster 3 (Urban/Anthropogenic/Transport) exhibits the highest CCN concentrations and maintains elevated values through much of the sampled depth. Within the PBLHT, CCN concentrations at SS = 0.80 % reach ∼ 3000 cm-3. The ratio (N0.80%/N0.13%) is about 5, suggesting a broader, but still relatively processed, size distribution. This is consistent with aged urban/industrial pollution, where the transport time from Houston likely allows for photochemical aging and coagulation, and secondary aerosol formation (sulfate, nitrate, SOA) has grown particles into the accumulation mode, but a fraction of smaller, less hygroscopic particles remains (Li et al., 2025; Farley et al., 2024). This interpretation is further supported by the characteristics of aerosol chemical composition. The presence of an “aged” or “processed” aerosol population was observed by chemical analysis of a 3 June sample representative of Cluster 3's influence. This sample was exceptionally complex, with 1020 molecular formulas (MFs) assigned (Fig. S6 in the Supplement). A large fraction of these were nitrogen- and sulfur-containing organics, including 340 organonitrates (CHON) and 104 organosulfates (CHOS). The prevalence of these compounds resulted in a distinct chemical signature: high oxygenation and low aromaticity, both hallmarks of extensive atmospheric processing. A detailed manuscript (under review) presents chemical analyses of organic aerosols collected during TBS flights on 3, 5, 8, and 9 June 2022, and reports significant day-to-day variability attributable to distinct air-mass influences.

Cluster 2 (Mixed Influence from marine and regional) represents air masses influenced by both marine and regional continental sources and sits between Clusters 1 and 3 in absolute CCN concentration but shows the highest spread ratio (N0.80%/N0.13% ∼ 6). Cluster 2 exhibits a distinct vertical structure compared with Clusters 1 and 3. In addition to being well-mixed in the lower 70 % of PBL, CCN concentrations influenced by Cluster 2 increase significantly with altitude at the top of the PBLHT (above 1000 m), culminating in a distinct peak near the PBL top (Altitude/PBLHT > 0.8) with values exceeding 5000 cm-3 at SS = 0.80 %. This accumulation of aerosols aloft is strong evidence of active in situ particle growth occurring preferentially at the top of the boundary layer. These continental pollutants are trapped beneath a temperature inversion and interact with the high humidity and strong solar radiation characteristic of the coastal marine environment. These conditions are ideal for efficient aqueous-phase chemistry (cloud processing) and photochemical reactions, which rapidly convert precursor gases into particulate matter and add mass to existing small particles, thereby growing them into the CCN size range (Li et al., 2025; Farley et al., 2024). However, Fig. 7d shows that the number of samples decreases markedly with altitude, so the upper-level enhancement in Cluster 2 should be interpreted with caution, and additional observations would be needed to confirm this as a systematic feature.

We noticed elevated CCN layers in Fig. S3c (1.5 < Altitude/PBLHT < 2.2), which is due to the single-day event on 9 September, might not represent the general trend in this airmass regime.

A central obstacle in observational aerosol–cloud interaction research is disentangling aerosol microphysical effects from the meteorological variability that simultaneously governs both aerosol distributions and cloud formation. The 6–7 September 2022, case study from the TRACER campaign near Houston provides a compelling, data-rich illustration of this challenge – demonstrating how mesoscale circulations, airmass advection, and differential transport can simultaneously transform the thermodynamic environment and the aerosol population, rendering causal attribution of cloud changes to aerosol effects deeply ambiguous from observations alone.