Aerosol properties exhibit pronounced vertical variability that strongly influences their radiative effects, interactions with clouds, and atmospheric lifetimes . Accurate characterization of the vertical structure of aerosols is therefore essential for understanding their impacts on weather, air quality, and climate . Aerosols within the planetary boundary layer (PBL) often differ substantially from those in the free troposphere in their optical, chemical, and microphysical properties . These contrasts lead to distinct radiative behaviours and cloud interaction potentials. Consequently, vertically resolved aerosol observations are indispensable for improving radiative transfer calculations, validating satellite retrievals, and constraining atmospheric models .

Vertical profiles of aerosols can be obtained using both in situ and remote sensing techniques. Recent advances in instrumentation have enabled increasingly detailed profiling of aerosol optical and microphysical characteristics . Ground-based and satellite-borne lidar systems have been widely employed to retrieve backscatter and extinction profiles . However, lidar observations are often constrained by signal attenuation in dense aerosol layers, operational challenges in complex or urban environments, and retrieval uncertainties associated with assumed aerosol types.

To validate the lidar observations, balloon-borne measurements could be used, such as those performed by Compact Optical Backscatter Aerosol Detector (COBALD) , a lightweight aerosol backscatter detector.

In situ measurements from aircraft campaigns provide the most direct and accurate characterisation of aerosol properties . Nevertheless, the high operational costs and limited spatial and temporal coverage of aircraft observations restrict their routine use. Alternative approaches using unmanned aerial vehicles (UAVs), tethered balloons , and other innovative platforms have gained increasing attention for profiling the lower troposphere and near-surface pollution layers . Further progress has been achieved through the use of cable cars , mobile urban systems , and measurement systems deployed on tall buildings or towers . Despite continued developments , most current systems remain constrained by payload capacity, vertical range, or operational complexity.

There is therefore a continuing need for cost-effective, flexible, and high-vertical-resolution platforms capable of routine in situ aerosol profiling. To address this gap, we developed AeroSonde, a novel radiosonde-based system designed to measure vertical profiles of aerosol optical and microphysical properties, with a particular focus on long-range transport of aerosol (e.g., biomass burning (BB), dust, volcanic). The system can operate with tethered balloons, UAVs, or free-flying radiosondes, enabling measurements across diverse environments-from urban pollution layers to remote BB regions. The primary objective of this study was to assess the AeroSonde's performance and demonstrate its capability for vertical aerosol profiling. For this purpose, initial field deployments were conducted at the Poland-AOD atmospheric observation site in Strzyżów (Poland) during episodes of elevated aerosol loading associated with long-range BB aerosol and mineral dust transport in summer 2025. In the absence of independent in situ reference measurements, remote sensing measurements were used to evaluate the optical properties retrieved from the AeroSonde.

The structure of this paper is organised as follows. Section  describes the AeroSonde, the set of instruments, and the measurement site. Section  outlines the calibration procedures and evaluates instrument performance. Section  presents the observational results, followed by a discussion of their implications in Sect. . The main conclusions and perspectives are summarised in Sect. .

The calibration and data evaluation process aimed to verify the consistency and accuracy of the optical sensors used in the study, as well as to assess their performance under real atmospheric conditions. Measurements were conducted at the Radiative Transfer Laboratory (RTL) of the Institute of Geophysics, University of Warsaw, over 22 d during the summer of 2025. Five SPS30 sensors were installed inside a meteorological shield to protect them from precipitation. The average Pearson correlation coefficient across all instrument pairs was 0.97 for PM₁₀ and PM₁, and 0.99 for TPS when 1 h averaging was applied. In contrast, for the original 1 s data (without averaging), the correlations decreased to 0.89 for PM₁₀ and 0.90 for PM₁. The mean relative differences were 8.9 % and 8.7 % for PM₁₀ and PM₁, respectively. Significantly larger differences (38 % for PM₁₀) were observed for the non-averaged data (1 s resolution), which corresponds to the statistical noise of the detectors. In the case of TPS, the absolute mean difference between all SPS30 pairs was 0.023 µm, corresponding to a relative difference of 4.5 %. These results demonstrate that the discrepancies among the readings of the low-cost SPS30 sensors were surprisingly small.

Additionally, two experiments were conducted to evaluate the instrument’s response time and its performance under low atmospheric pressure conditions. In the first experiment, the sensor was placed in a chamber with a very high aerosol concentration (PM₁₀ in the order of 200–300 µgm-3). The sensor's response to a sudden change in conditions showed that the time constant did not exceed a few seconds, typically being less than 5 s. For typical balloon ascent rates of 3–4 m s⁻¹, this corresponds to a vertical resolution of no worse than 15–20 m. The sensor's performance under reduced pressure was tested in a laboratory vacuum chamber, modified to enable fan speed (revolutions per minute, RPM) measurements over a pressure range from 1000 to 50 hPa. One of the three wires of the SPS30's flow-inducing fan – corresponding to the tachometric (tachometer) output – was connected to an Arduino board to count pulses associated with fan rotations. The results indicated that the fan speed remained stable at approximately 6000 RPM throughout the experiment (see Fig. S1), confirming that SPS30 airflow does not require correction for altitude (pressure) effects.

The low-cost optical particle sensors (SPS30) were calibrated against reference-grade aerosol instrumentation. The ASC and the SAE were referenced to an Ecotech Aurora 4000 integrating nephelometer, while the Reff was derived by comparison with the TSI 3340 LAS. For the ASC derived from the SPS30 sensors, a non-linear regression model was applied to convert particulate mass concentration to ASC, expressed as: 1ASC=a⋅PM10+b⋅PM10, where a and b are empirical calibration coefficients determined by non-linear least-squares fitting. The proposed relationship between ASC and PM₁₀, specifically the inclusion of both a linear and a square-root term, is supported by simulations conducted as part of this study. Lorenz-Mie simulations for spherical particles (Fig. S2) show that this functional form provides a good approximation of both the simulated and observed data, effectively capturing the combined effects of particle number concentration and particle size evolution. This suggests that the non-linear regression reflects the underlying physical processes governing aerosol behavior, including changes in particle size due to hygroscopic growth and coagulation. This interpretation is consistent with the findings of , who showed that mass scattering efficiency (i.e., the ASC to PM₁₀ ratio) is a non-linear function of particle size. However, if the particle size distribution remains constant and only the particle number concentration varies, a linear relationship between ASC and PM₁₀ would be expected. Figure a presents a scatter plot of PM₁₀ from the SPS30 sensors versus ASC from the Aurora 4000 at 525 nm. Several data points deviating from the regression curve at PM₁₀ concentrations of approximately 10–20 µgm-3 were associated with a single atypical aerosol episode. During this episode, aerosol transport models indicated the presence of sea-spray aerosol over Warsaw, located approximately 300 km from the Baltic Sea coast. The observed deviations may therefore be related to differences in aerosol type and particle properties during this episode, which can modify the relationship between PM10 mass concentration measured by the SPS30 and aerosol scattering measured by the Aurora 4000. Nevertheless, a very high coefficient of determination (R2 = 0.96) indicates good agreement between PM₁₀ and ASC from the Aurora 4000. The root mean square error (RMSE) is 5.1 Mm⁻¹, corresponding to only 3.8 %. These results are consistent with previous studies , which demonstrated that optical low-cost sensors can reliably measure ASC. The SAE from the SPS30 sensors was calibrated against reference values obtained from the Aurora 4000 nephelometer using a second-order polynomial regression based on the SPS30’s typical particle size (TPS): 2SAE=c0+c1⋅TPS+c2⋅TPS2, where TPS represents an optical parameter derived from the SPS30 sensor, and c0, c1, c2 are the fitted coefficients. According to the manufacturer, TPS is related to the characteristic mean particle size of the sampled aerosol and correlates with the weighted average of the number concentrations measured in the particle-size bins. However, the detailed algorithm used to derive TPS is proprietary and is not disclosed by the manufacturer. Therefore, the polynomial relationship was determined empirically from the calibration data rather than derived from an assumed analytical relationship between TPS and SAE. Despite this uncertainty, the polynomial fits provide reasonable approximations, capturing the general trends observed in the data. This non-linear behavior is consistent with previous studies, which have shown that key optical parameters, such as the Ångström exponent, are non-linearly related to particle size . Nevertheless, further studies are needed to refine these relationships and better understand the underlying physical processes. The agreement between SAE from the SPS30 and the Aurora 4000 reference is noticeably weaker (Fig. b; R2 = 0.78, RMSE = 0.24). This reflects both the limitations of the SPS30 in retrieving this parameter and the relatively high uncertainty of SAE even in professional nephelometer measurements. Nevertheless, the SPS30 can still be used to distinguish particle types characterized by different size distributions – for example, BB aerosol versus mineral dust or anthropogenic pollution . Similarly, the Reff was estimated from the SPS30 TPS values by comparison with the LAS (TSI 3340) measurements. A quadratic regression model was applied: 3Reff=d0+d1⋅TPS+d2⋅TPS2, where TPS represents an optical parameter derived from the SPS30 sensor, and d0, d1, d2 are the fitted coefficients. As in the case of SAE, the Reff can be estimated, albeit with relatively high uncertainty (Fig. c; RMSE = 0.019 µm). Table  summarizes the calibration statistics for all five SPS30 sensors against the Aurora 4000 nephelometer and the LAS instrument. The consistently high coefficients of determination confirm the strong agreement between the SPS30 sensors and the reference instruments. In contrast, the calibration of SAE and Reff is meaningful only when measurements include a sufficiently broad range of aerosol size conditions - particularly during episodes dominated by coarse-mode particles such as mineral dust or marine sea spray. Such conditions were present in our dataset, as indicated by elevated TPS values reported by the SPS30 sensors. In contrast, datasets without these episodes (with TPS typically ranging between 0.4 and 0.55 µm) provide too limited a dynamic range to enable reliable calibration of SAE or Reff.

The uncertainties of defined parameters (ASC, SAE, Reff) are calculated from the calibration curve error (Table ) and the reference instrument (Aurora 4000 and LAS) error 4δASC=δfitASC2+δASCAurora2, where δfitASC is uncertainty of ASC calibration fit (Eq. ) and δASCAurora is the uncertainty of Aurora ASC. For the remaining two quantities (SAE and Reff), an analogous equation applies.

The total uncertainty of the ASC measured with the Aurora 4000 was estimated by combining relative and absolute uncertainty components. Based on previous intercomparison studies and manufacturer data , typical values of the relative uncertainties of individual components are span calibration (2.5 %), zero check (2 %), illumination/cosine response (5 %). In case of the truncation correction, the uncertainty depends on particle size and for fine particles is about 5 %, while for coarse mode it can be up to 33 % . Therefore, this term is estimated based on SAE as follows: 5δASCtrunc=0.05+0.281-SAE/2.0. Assuming these errors are statistically independent, the combined relative uncertainty was estimated using the root-sum-square method. For the mean values listed above, the resulting total relative uncertainty of ASC is approximately 14 %. Since such relative uncertainty produces unrealistically small values of absolute uncertainty during very clean conditions, the absolute error is added. It accounts for instrumental baseline drift, zero offset, and electronic noise. We assume that this uncertainty is about 1 Mm⁻¹. The total uncertainty δASCAurora of the scattering coefficient was thus calculated as: 6δASCAurora=δcal2+δzero2+δtrunc2+δcos2+δabs2. where δcal,δzero,δtrunc,δcos,δabs are the uncertainties of span calibration, zero check, truncation error, cosine response, and absolute calibration, respectively. The uncertainty of the SAE was then derived using standard propagation of uncertainty based on the SAE at two wavelengths (450 and 635 nm): 7δSAEAurora=1|ln⁡(λ1/λ2)|δASCAurora,λ1ASCAurora,λ12+δASCAurora,λ2ASCAurora,λ22.

This approach provides a realistic estimate of the total uncertainty in both ASC and the spectral dependence of aerosol optical properties (SAE). The absolute uncertainty for SAE increases significantly for small ASC. In the case of the LAS instrument, the uncertainty of Reff was estimated during two 2-week campaigns in Sopot (Poland) in 2023. Comparison of LAS to NanoScan SMPS Nanoparticle Sizer spectrometer and Optical Particle Sizer (OPS) 3330 devices allows us to estimate the Reff uncertainty as 0.028 µm. Table  summarizes the measurement uncertainties of ASC, SAE, and Reff under different optical conditions. The uncertainty in ASC shows only a weak dependence on pollution level, increasing slightly from approximately 35 % at ASC = 10 Mm⁻¹ to about 39 % at ASC = 200 Mm⁻¹. For SAE, the uncertainty decreases modestly with increasing ASC and declines more markedly with increasing SAE. In contrast, the uncertainty in Reff exhibits only a weak dependence on its magnitude. Overall, the uncertainties associated with SAE and Reff are substantial, indicating that these parameters can be retrieved from SPS30 measurements only with limited accuracy. For instance, for a typical SAE value of 1.50, the 95 % confidence interval spans nearly the full range of observed SAE variability.

The PM_2.5 concentration is calibrated by the manufacturer against a TSI DustTrak DRX 8533, whereas PM₁₀ is derived from the particle-size distribution measured by the SPS30. However, no independent field calibration of the PM_2.5 and PM₁₀ mass concentrations was performed in the present study, as no reference-grade PM mass instrument was available during the campaign. For reference, the manufacturer specifies a mass concentration precision of ±(5 µgm-3 + 5 % of the measured value) for PM_2.5 and ±25 µgm-3 for PM₁₀ in the concentration range of 0–100 µgm-3. These specifications represent the manufacturer-reported between-device variation rather than an uncertainty estimate established for the present measurements.

The AeroSonde system was tested in summer 2025. Three observational campaigns were conducted, during which five vertical profiles were obtained – for details see Table . Measurement dates were selected based on meteorological conditions favorable for remote sensing observations, as well as the occurrence of long-range transport episodes in the middle and upper troposphere. During the first two campaigns (9–10 June and 7 August), transport of BB aerosols from North America was observed, while the final campaign (29 August) captured a Saharan dust transport event. The mean ascent rate of the AeroSonde ranged from 2.7 to 3.9 m s⁻¹, reaching altitudes from 10 to 26.8 km. Such vertical velocity corresponds to a vertical resolution of at least 15–20 m for the SPS30 sensor.

Due to strong wind speeds in the middle and upper troposphere, the AeroSonde experienced significant horizontal displacement (from approximately 75 to 120 km), mainly toward the east (BB case) and northeast (dust case, see Fig. a). The balloon trajectories are approximately linear. Only for the highest trajectories, a change in direction is observed.

Figure b shows the 3D trajectories of the AeroSonde platform carried by a balloon, highlighting the vertical variability of the ASC along the profiles.

The color-coded ASC values show a clear tendency toward higher values in the lower and middle troposphere. With increasing altitude, ASC generally decreases, except for the profile acquired on 9 June, where an increase in ASC is observed near 10 km. However, distinct elevated layers with enhanced ASC are intermittently observed, suggesting the presence of aerosol stratification and long-range transport features. These layers exhibit variability both within individual profiles and between different measurement days.

Due to the significant horizontal displacement of the sensors suspended beneath the balloon from the station, the methodology described in Sect. 2.3.2 was applied to get the AEC profiles. Comparison between profiles of ASC retrieved from AeroSonde measurements and AEC profiles obtained using the Klett LR method (AEC_LR; Fig. a) and the Klett AOD approach (AEC_AOD; Fig. b) was performed using 100 m vertical averages. Comparison was done for all available measurements, aggregated over all campaigns. For AEC_LR, ASC exhibits a Pearson correlation of r = 0.81 with an RMSE of 18.5 Mm⁻¹, while for AEC_AOD, the correlation is slightly lower (r = 0.74) with a higher RMSE of 25.1 Mm⁻¹. Both lidar retrievals show good agreement with AeroSonde-derived ASC, with the Klett LR method providing closer correspondence. The mean bias is positive in both cases-3.9 Mm⁻¹ for AEC_LR and 6.4 Mm⁻¹ for AEC_AOD-which is consistent with the expectation that AEC exceeds ASC due to aerosol absorption.

Figure  shows the distributions of differences between AEC and ASC for all available measurements, aggregated over all campaigns. In the lowest layer (0.7–2 km) (Fig. a, b), both retrieval approaches exhibit a positive bias, which is larger for the AOD-constrained method (BIAS = 18.2 Mm⁻¹, RMSE = 25.6 Mm⁻¹) compared to the LR-based approach (BIAS = 12.5 Mm⁻¹, RMSE = 16.3 Mm⁻¹). The mean ASC in this layer is 20.5 Mm⁻¹, while the AEC is 33.0 and 38.7 Mm⁻¹, respectively, for the approach based on LR and AOD.

In the 2–5 km range (Fig. c, d), the distributions become broader for both lidar methods, indicating larger variability in AEC-ASC differences, with higher BIAS and RMSE for the AOD-based retrieval (BIAS = 12.7 Mm⁻¹, RMSE = 41.9 Mm⁻¹) than for the LR-based approach (BIAS = 8.0 Mm⁻¹, RMSE = 28.4 Mm⁻¹). In this layer the highest values of ASC (34.0 Mm⁻¹) and AEC (42.0 and 46.7 Mm⁻¹) are observed, indicating long-range transport of aerosols above the PBL.

At higher altitudes (5–8 km) (Fig. e, f), both methods show near-zero bias and narrower, more symmetric distributions, with BIAS = -1.9 and -0.6 Mm⁻¹ and RMSE = 17.9 and 17.5 Mm⁻¹ for the LR and AOD approaches, respectively. The slightly negative bias is somewhat unexpected, as AEC is generally expected to exceed ASC; however, the magnitude of this effect remains small. The mean ASC in this case is similar to mean between 0.7 and 2 km.

Above 8 km (Fig. g, h) both approaches also exhibit minimal bias (BIAS = 0.4 and 1.4 Mm⁻¹) and low RMSE (5.2 and 7.0 Mm⁻¹). In this case, the mean ASC is 8.0 Mm⁻¹, while AEC is 8.5 and 9.4 Mm⁻¹, respectively, for LR and AOD method.

Vertical profiles obtained using the AeroSonde and lidar show a similar aerosol layer structure in both qualitative and quantitative terms. Some differences in the values of optical parameters are evident and arise from several factors. First, they result from the different physical quantities measured by the two instruments (ASC versus AEC). Second, they are related to the measurement limitations of the SPS30 sensor and the inversion methods applied to lidar data in the elastic 532 nm channel. Third, they stem from the fact that the balloon-borne sonde does not follow a purely vertical path, and atmospheric conditions vary along its trajectory. Figures  and  show that the altitudes of layers with maximum values can differ by up to 100 m between AeroSonde and lidar measurements. Averaging over 100 m vertical intervals mitigates discrepancies arising from the differing sampling strategies of the two systems, with lidar probing a fixed vertical column and the AeroSonde following a slanted ascent profile. The assumption of horizontal homogeneity of the wind field, applied to enable lidar-sonde comparisons (Appendix ), is a simplification of real atmospheric conditions and may introduce additional pointwise differences between AEC and ASC at individual altitudes. It should be noted that previous studies primarily focused on the validation of lidar-derived aerosol properties (backscattering coefficient) using in situ measurements . Therefore, in the present work, lidar observations cannot be treated as an absolute reference, but rather as an independent dataset for comparison.

The vertical profiles of ASC obtained from the AeroSonde enable estimation of SAOD, columnar SAE, and, when combined with AERONET single-scattering albedo (SSA) data, also AOD. Table  compares these quantities with measurements from the sun photometer. The AOD comparison shows excellent agreement on 9 June (bias 0.01), moderate discrepancies on 7 August (0.05), and the largest difference on 29 August (0.11). During the Saharan dust intrusion, both AOD and ASC measured by the AeroSonde were significantly underestimated relative to AERONET and lidar-derived AEC.

Several factors contribute to these discrepancies. First, mineral dust can substantially increase AEC due to absorption , with reported single scattering albedo values typically ranging from approximately 0.85 to 0.95 depending on composition and wavelength . This indicates a non-negligible absorption contribution, particularly in the presence of iron oxides, which may lead to underestimation of extinction when using scattering-based measurements alone. Second, the use of only elastic lidar channels limits retrieval accuracy; Raman channels would provide more precise characterization , but were not available during these measurements. Third, particle nonsphericity and differing measurement geometries affect the comparisons: the SPS30 measures at a scattering angle of approximately 90°, whereas the lidar detects backscatter at 180°. For mineral dust, the scattering phase function reaches a minimum near 90°, causing the SPS30 to potentially underestimate ASC . It should also be noted that no depolarization measurements were available, as the lidar system operated at a single wavelength without polarization capability. Such measurements would enable the identification of nonspherical particles, such as mineral dust, and could therefore provide additional constraints to better interpret the observed discrepancies.

Previous studies indicate that the SPS30 performance decreases in the presence of coarse, dust-like aerosols, particularly for PM₁₀ . Field observations during Saharan dust events show that PM₁₀ is often underestimated relative to reference instruments, reflecting the sensor's design, which is better suited for fine-mode aerosols . Laboratory and intercomparison studies further demonstrate limited size selectivity in low-cost optical sensors, including the SPS30, with coarse particles (> 5 µm) prone to misclassification or undercounting due to restricted bin resolution and reliance on optical scattering principles . Moreover, laboratory experiments with monodisperse aerosols reveal that the SPS30 response depends on particle size and material properties, with increasing deviations for larger particles .

These results highlight the need for aerosol-type-dependent calibration and careful interpretation of SPS30-derived PM₁₀ data, particularly when relying on factory calibration in dust-influenced environments . To address this limitation, the SPS30 should be calibrated against the Aurora 4000 under a range of aerosol conditions, capturing the variability in the conversion between particle mass (as reported by the SPS30) and optical properties arising from differences in chemical composition and particle size. Accordingly, Eq. () should be revised to explicitly account for the relationship between reported mass concentration and ASC.

Differences in measurement geometry may also contribute to discrepancies between AeroSonde, lidar, and sun photometer columnar properties. During the Saharan dust event, the balloon-borne sonde travelled northeast, the lidar performed vertical profiling, and the sun photometer measured AOD along a westward slanted column. Given the strong spatial and temporal variability during this event, these differing measurement geometries could have introduced substantial variability in the retrieved optical quantities .

The uncertainty of AeroSonde-derived ASC ranges from 35 %–43 %, while lidar AEC uncertainties vary between 10 %–35 % (Klett LR) and 2 %–19 % (Klett AOD). Derived quantities such as SAE and Reff exhibit larger uncertainties, particularly for clean layers with low ASC; these uncertainties amount to approximately ±1.0 for SAE and range from ±0.05 to ±0.1 µm for Reff. Columnar quantities (SAOD, SAE_col, Reff,col) were evaluated assuming full correlation of uncertainties with altitude, yielding upper-limit uncertainties of 33 %–43 % for SAOD and ±1.0 for SAE_col and ±0.07 µm for Reff,col. Comparisons with AERONET indicate that SAE retrievals from AeroSonde are less robust and should be interpreted cautiously.

A rigorous quantification of uncertainties in the column-integrated parameters (SAOD and SAE_col) remains challenging, as the vertical correlation of uncertainties in ASC and SAE is unknown. The maximum uncertainty in SAOD is estimated by assuming full correlation (correlation coefficient equal to 1) between ASC uncertainties at all altitude levels. Similarly, the upper-limit uncertainty of SAE_col is calculated under the assumption of full correlation between ASC and SAE uncertainties at each altitude. These estimates, therefore, represent conservative upper bounds rather than precise uncertainty quantification, since a realistic characterization of vertical error correlations is not feasible in the present study. Under these assumptions, the upper limit of SAOD uncertainty ranges from 33 % to 43 %.

In contrast, uncertainties in SAE_col and Reff,col are substantially larger, exceeding ±1.0 for SAE_col and ±0.06–0.07 µm for Reff,col. For AERONET AE_440−870, reported uncertainties are on the order of ±0.3 at AOD₄₄₀ ∼ 0.15, increasing to ±0.6 when AOD₄₄₀ drops to ∼ 0.08, showing an approximate inverse dependence on AOD . This comparison further indicates that SAE retrievals from AeroSonde data are less robust and should be interpreted with caution, particularly when used for aerosol-type classification.

AeroSonde measurements also enable estimation of the contribution of the PBL to total AOD. During the analysed long-range transport events, the contribution from aerosol found within PBL ranged from approximately 1 % to 36 %, which is comparable to lidar-based estimates that varied between 1 % and 33 % for the Klett AOD retrieval and between 1 % and 39 % for the Klett LR method (Table ). These findings are broadly consistent with CALIOP aerosol extinction data, indicating that roughly 69 % of tropospheric AOD originates within the PBL and 31 % in the free troposphere, with a similar day-night vertical distribution due to the persistence of a residual layer at night . Regional lidar studies in Warsaw show that aerosols within the PBL contribute approximately 61 % of total AOD . Unlike lidar measurements, the AeroSonde approach is not affected by limitations associated with PBL heights below the lidar overlap layer, enabling more reliable characterization of boundary-layer aerosol contributions.

Compared to UAV-based measurement platforms, the balloon-borne AeroSonde offers several distinct advantages that enhance its suitability for atmospheric aerosol studies. Its passive ascent, without active air intake, prevents flow-induced artefacts and avoids particle size-selective sampling that can result from propeller-induced turbulence in drones, enabling a more representative characterization of the ambient aerosol population. Additionally, the AeroSonde can reach altitudes extending into the stratosphere, making it particularly valuable for monitoring phenomena such as volcanic eruption plumes, whereas UAV platforms are generally restricted to the lower troposphere. A notable limitation of the AeroSonde, however, is its typical non-recoverability after launch. In contrast, UAV systems allow sensors to be reused, facilitating repeated flights at lower operational cost. To ensure accurate measurements, each SPS30 sensor deployed on the AeroSonde must therefore be calibrated through colocated measurements with the Aurora 4000 across a wide range of aerosol conditions, capturing the full variability of both ASC and SAE.

Finally, the AeroSonde provides a valuable tool for validating vertical aerosol profiles from reanalysis and model products (e.g. CAMS, MERRA-2, NAAPS). Direct measurements of light scattering in the lower troposphere enable an independent assessment of modelled aerosol profiles.

Future work could expand the capabilities of the AeroSonde system. The measurement capability of the system could be extended by integrating additional sensors, subject to the mass constraints of the platform (currently 175 g including the housing). A single-channel aethalometer would allow the AAC, and hence the SSA, to be retrieved directly rather than relying on columnar AERONET values, which would substantially improve the AOD estimates discussed above. A dedicated optical particle sizer with extended size range would address the principal limitation identified in the dust case, namely the reduced sensitivity of the SPS30 to coarse-mode particles. A further possibility is closer integration with the meteorological radiosonde, so that thermodynamic and aerosol measurements are provided by a single unit. In each case, the added mass and power consumption must be weighed against the benefit.

Beyond the platform itself, combining the AeroSonde with a Raman lidar would allow retrieval of AEC without assuming a fixed LR or relying solely on AOD, improving the quantitative accuracy of vertical aerosol characterisation and enhancing closure studies.

This study presents and demonstrates a new balloon-borne measurement system (AeroSonde) designed to provide high-resolution, vertically resolved observations of aerosol optical properties, delivering profiles from the surface up to the upper troposphere and, in some cases, the lower stratosphere, with a vertical resolution of a few meters. The performance of the AeroSonde was evaluated across a limited number of case studies, which should be regarded as a proof-of-concept of the system’s capabilities. The measurements enable the estimation of ASC profiles, SAOD, and AOD, and show potential for supporting the evaluation of reanalysis products and satellite-based aerosol retrievals. The results indicate generally good agreement with lidar and AERONET observations, within the expected uncertainty range, and demonstrate that the SPS30 low-cost sensor is capable of capturing the main features of aerosol vertical structure. In particular, the system allows the identification of boundary-layer contributions and elevated layers associated with long-range transport. However, parameters such as SAE and Reff should be interpreted with caution and are better suited for qualitative characterization.

The most important findings are:

The vertical profiles of ASC measured by AeroSonde are consistent with simultaneous lidar observations, with Pearson correlation coefficients of 0.74–0.81 and RMSE values of 18.5–25.1 Mm⁻¹, confirming the system's capability to resolve aerosol layering and optical properties.

Despite its limitations, the AeroSonde system demonstrates clear potential as a complementary tool to ground-based remote sensing and traditional radiosonde observations. Future work should focus on improved calibration strategies, particularly for coarse-mode aerosols such as mineral dust and sea salt, which remain a known limitation of low-cost optical sensors. Additional validation against reference-grade instruments is also required to better quantify measurement uncertainties, including the PM_2.5 and PM₁₀ mass concentrations, which were not independently calibrated in the present study. Validation covering a range of aerosol types and size distributions would allow a more robust assessment of performance of sensors and measurement uncertainty under atmospheric conditions. Furthermore, extended deployments under a wider range of aerosol regimes would help to assess the robustness of the proposed approach. The additional integration of the AeroSonde system with a Raman lidar would provide complementary information by enabling the retrieval of aerosol extinction profiles without assuming LR or AOD, which are typically required when using conventional lidar alone. Raman lidar, in particular, enables the determination of AE, which can be compared with the data obtained from the AeroSonde, offering deeper insight into the particle size distribution. Moreover, combining Raman lidar's AEC with the AeroSonde's ASC would allow estimation of the single scattering albedo. This combination would not only improve the accuracy of vertical aerosol profiles but also provide more reliable data for closure studies, as the dependence on assumptions such as LR and AOD would be minimized. These developments would further improve the quantitative capabilities of the AeroSonde system and broaden its applicability in atmospheric aerosol research.