The sampling platform (Fig. ) was developed using the DJI Matrice 350 drone (SZ DJI Technology Co., Ltd., Shenzhen, China). The fully equipped system, including sensors and a sampling setup (approximately 980 g), resulted in a total flight weight of approximately 7.45 kg, allowing for a flight time of 30 min, including battery redundancy for safe landing. The platform was equipped with a sonic anemometer for wind speed and direction (TriSonica Mini (LI-COR Environmental, Lincoln, NE, USA)), sensors for T and relative humidity (RH) (SKS21 (Sparv Embedded, Linköping, Sweden)), and pressure (SKH3 (Sparv Embedded, Linköping, Sweden)). The anemometer was positioned approximately 2.5 rotor radii (80 cm) from the rotors to reduce propeller-induced turbulence while ensuring structural feasibility. Wind speed and direction were measured at 5 Hz using the sonic anemometer, with a resolution of 0.1 m s⁻¹. The T sensor, reporting a resolution of 0.1 °C with a 0.32 s response time. The RH sensor has a resolution of 0.1 % RH with a response time of 0.5 s at 23 °C. Both T and RH sensors were mounted 20 cm from the drone body to minimize interference from onboard heat sources All meteorological data was integrated into the Sparvio (Sparv Embedded, Linköping, Sweden) system and logged at full resolution.

The particles were collected on a 25 mm diameter Mitex™ PTFE filter with a stainless steel filter holder (Sartorius, Germany) via a 40 cm, 4 mm inner diameter stainless steel inlet co-located with the anemometer to minimize spatial lag and avoid potential interference from rotor downwash. In this configuration, no solar radiation shield was used as all tests were made in Nordic cold to mild temperatures with limited direct solar radiation. When operating in warmer environments, the use of a shielded filter holder may be necessary to minimize artifacts related to thermal effects. A comparison of online FIGAERO-CIMS and drone-mounted filter holders is shown in Fig. . Aerosol sampling was conducted using a pump (KVP8-KB, Kamoer, China) mounted on a spring-damped PETG 3D-printed mount. Due to restriction by the sampling line and filter, the maximal sampling flow is 4.5 L min⁻¹ with 7 mm diameter effective sampling area and 8.5 L min⁻¹ with 25 mm effective sampling area. We have also used a Leland Legacy pump (SKC Ltd., Dorset, UK), which offers 4.2 L min⁻¹ with 25 mm sampling area. The current application of the platform focuses on bulk characterization, thus no size-selective inlet is applied and the system nominally sampled total suspended particulate (TSP). However, estimations of the transmission efficiency of the inlet indicated that larger particles (10 µm in diameter) may suffer losses exceeding 80 %, while particles smaller than 2.5 µm are expected to have a transmission efficiency over 95 %. In the present platform configuration, these losses primarily occur at the inlet interface, and are largely independent of the downstream inlet architecture. Consequently, similar inlet conditions are expected for the ground-based sampling configuration used for comparisons. The platform design nevertheless allows straightforward integration of size-selective inlets for studies targeting specific research objectives, such as size-resolved chemical composition or health-relevant aerosol fractions.

In the online sampling system, i.e. CIMS coupled with the FIGAERO inlet , particles were collected on the same 25 mm PTFE filter, and restricted within the central 7 mm diameter area by a Teflon™ insert (25 mm OD with 7 mm central through-hole) (Fig. b). To ensure uniformity in collection efficiency, a similar Teflon™ insert or an aluminum foil (both 25 mm OD with 7 mm central through-hole) was applied on the top of the filter, matching the effective collection and desorption area of the online FIGAERO system (Fig. ). Real-time telemetry enabled adaptive sampling and sensor status monitoring. Pump operation was remotely controlled, and activation times were logged. The entire system was powered by the drone's internal battery, avoiding the need for external sources and ensuring uninterrupted operation. Using regulated onboard power also ensures the pump flow rate remains stable during the duration of the flight.

For effective online and offline sampling, sufficient particle mass must be collected on the filter, which depends on the ambient particle concentration, and sampling volume (sampling duration × sampling flow rate). In clean environments, higher sampling volume is required, compared to urban and semi-rural settings which typically offer higher mass loading . The typical sampling flow of the online FIGAERO-CIMS is 4–10 L min⁻¹. With 5 L min⁻¹ under ambient aerosol concentrations of 5 µg m⁻³, the configuration yields a collected mass of 0.75 µg over a sampling duration of 30 min. Depending on environmental conditions, the sampling duration can be extended, e.g. to 2–3 h.

When transferring the method to drone platforms, additional constraints arise from payload capacity and flight endurance. In such cases, pump capacity, i.e maximal sampling flow rate, cannot be indefinitely scaled due to weight limitations. The sampling flow of our system, as mentioned above, is 4.5 L min⁻¹ and the collection time could be extended by utilizing a remote-controlled pump that allows pausing sampling during battery swaps. A full stop, landing, battery replacement, and re-ascent cycle can be completed in a few minuntes depending on the sampling altitude and desired flight paths, enabling nearly continuous hour-scale collection. When filters are analyzed semi-online, i.e. immediately following collection, sampling durations can be tailored dynamically based on signal strength detected by the CIMS or the organic mass concentration detected by ancillary instrument, such as an aerosol chemical speciation monitor (ACSM). Remote pump control and telemetry enabled strategic spatio-temporal sampling, allowing drone-based filters to be positioned precisely in horizontal or vertical alignment with ground-based systems and meteorological features. After sampling, filters are either transferred directly from the sampler to the FIGAERO inlet for analysis or stored in Petri dishes, wrapped in aluminum foil, and kept at 0 °C in a sealed freezer box until analysis (normally within 1–3 h for semi-online mode).

For offline analysis, the same sampling procedure is applied; however, chemical analysis using the FIGAERO-CIMS is conducted in the lab days to months after sampling collection. While the semi-online mode enables near-real-time characterization of particle chemical composition and allows direct comparison with stationary online measurements, the offline approach increases operational flexibility as it eliminates the need to deploy the mass spectrometer in the field or in a nearby laboratory. However, the offline approach may introduce potential artifacts associated with the potential condensation, re-evaporation, and reactions on the filter during storage, which may result in both positive and negative sampling biases .

FIGAERO-CIMS provides chemical molecular characterization of aerosol particles. It operates on the principle of filter pre-concentration followed by thermal desorption to enable ion-molecule adduct (here iodide as reagent ion) formation of particulate constituents. In online mode, simultaneous filter collection and gas-phase analysis allows for pseudo-continuous, parallel gas- and particle-phase measurements .

Adapting FIGAERO-CIMS to offline mode (only particle-phase) involves collecting particles onto filters that are subsequently placed in the FIGAERO filter holder for thermal desorption (Fig. ). This approach has been previously characterized and demonstrates good analytical reproducibility . Filter collected by the drone can be analyzed at will, either in a semi-online mode (immediately after sampling) or in an offline mode, i.e. at a later stage.

The analysis procedure using FIGAERO-CIMS is identical for both semi-online and offline modes. Filters undergo a desorption cycle: a 20 min T ramp from room T to 200 °C, followed by a 20 min soak at 200 °C, and a 12 min cooldown. When collecting two simultaneous filters (e.g. one from the drone platform and one from the ground-based sampler), the second is offset by 52 min, i.e desorption of the second filter is started immediately following the first. FIGAERO-CIMS data was processed using the Tofware software package (ToFware 4.0.3, Aerodyne Research Inc., Massachusetts, US) within the Igor Pro v9.0.5. environment (Wavemetrics, Inc., Oregon, US). From the filters collected during this campaign, 303 peaks were assigned at a mass resolution of ∼ 4000 m/Δm. Instrument background was quantified by running clean filters before each analysis segment and was subsequently removed individually for all identified compounds. Then, the background corrected thermograms were integrated, generating the total amount of ions for the full T ramp and soak (first 40 min of the cycle). This integrated signal was then normalized by the sampling volume for further comparison. In addition, the desorption T at which a compounds signal reaches its maximum (Tmax⁡) was extracted from the thermograms for each assigned peak . A detailed description of the instrument operation and data analysis is provided in our previous studies .

Field measurements (Table S1 in the Supplement) were performed in Gothenburg, Sweden during April 2024 and April to May 2025, Sweden. Gothenburg is a metropolitan coastal city with relatively low particulate loading (annual PM2.5<5 µg m⁻³ in 2024). Emissions are influenced by traffic, nearby industries, and port activities , while surrounding greenery provide biogenic VOCs. All measurements were made at the University of Gothenburg Natrium Atmosphere and Climate measurement site (Nat-AC-MS) located 100 m above sea level (m a.s.l.) and 25 m above ground level (m a.g.l.) in the central southwest of Gothenburg.

The drone platform's ability to perform accurate profiling using the onboard meteorological sensors was evaluated and characterized by comparing to the stationary readings (WXT210 Vaisala Oyj, Helsinki, Finland) at the Nat-AC-MS (Fig. S1). The drone platform hovered parallel to the weather station with intermittent landings to swap batteries. Wind speed, wind direction, T, and RH readings were averaged to 5 and 10 s and compared to assess the influence of the drone's rotor-induced turbulence on collection of meteorological data. As the averaging time dictates the spatial resolution of a vertical sounding at a fixed speed, the reduction in uncertainty (i.e. standard deviation) with longer averaging times has implications on the flight pattern. Furthermore, multiple vertical profiles at a speed of 1 m s⁻¹ and up to a height of 120 m a.g.l. were conducted during both day and night to assess the drone platform's capability to capture dynamic changes.

To evaluate the possible effects of rotor-induced flow on collection efficiency, filters were collected using the drone-based setup in parallel to a ground-based one. The parallel ground-based sampling system used the same filter holder as the drone-based system and was deployed on the Nat-AC-MS tower. During the first experiment both setups were sampling at a height of 8 m above ground (hovering conditions) and the drone was 10 m horizontally from the tower (Fig. S1) whilst during the second experiment they were sampling at a height of 2 m above ground (grounded conditions, inactive drone rotors) (see Table S1). These comparisons were made during relatively low (PM_2.5 ∼ 2–4 µg m⁻³) ambient aerosol concentrations. Drone and ground-based filters were both sampled at 4.5 L min⁻¹ for 3 h, resulting in a particle mass loading of at least 1.6–3.2 µg at an estimated PM2.5<2–4 µg m⁻³. Ambient particle loading was estimated using the publicly available City of Gothenburg monitoring program mass concentration measurements taken in the city center of Gothenburg (TEOM Thermo Fisher Scientific Inc., Waltham, MA, USA). The collected filters were analyzed immediately after sampling.

To showcase the capability of the drone platform to map a dynamic lower boundary layer, we present measurements from a typical early spring night characterized by a strong inversion. Vertical profiles of T, RH, wind speed and wind direction as well as aerosol samples were collected. The filter sampling period was a combination of three 30 min flights, with a total ∼ 1.5 h sampling time. Both ground and drone-based filter collections were started simultaneously when the drone reached 120 m a.g.l., and were intermittently stopped during descent and re-ascent periods to swap drone batteries. The drone-based filters were analyzed immediately by the FIGAERO-CIMS, followed by the filter collected using the ground-based sampling system. Vertical profiles of T, RH, and wind speed and direction were logged during the drone's ascent to 120 m at a speed of 1 m s⁻¹. Two supplementary vertical profiles of meteorological parameters were obtained before and after the full particle sampling sequence, to a total of 5 profiles. Here, a 5 s average is adopted for data collected vertically at a speed of 1 m s⁻¹ which results in a resolution of 5 m.

The wind speed and direction measured by the drone platform and the stationary measurements show good agreement over the testing period. Wind speeds during the sampling period ranged from of 1–7 m s⁻¹. The comparison between wind speed measured by the drone during hovering and that from the stationary sensor has a slope of 1.09 and an intercept of 0.3 m s⁻¹ (Pearson's correlation coefficient (R) 0.91, root mean square error (RMSE) 0.78 m s⁻¹) based on 5 s averages (Fig. ). At a 10 s average Pearson's R is increased from 0.91 to 0.93 and RMSE is reduced to 0.74 m s⁻¹, the intercept is reduced to 0.2 m s⁻¹. The overall uncertainty of azimuth and elevation angle is quantified with an RMSE of 11.3 and 7.7° and intercepts of 3.5 and 4.3° respectively. At a 10 s average, the RMSE is reduced to 8.6 and 6.6°, respectively.

As reported in previous studies bias in wind measurements is mainly caused by interference from the rotor down-flush, which is dependent on the drone platform setup, specifically the distance between rotor horizontal plane and anemometer, as well as wind speeds and drone movement, previously characterized as a function of rotor throttle or pitch angle. With a similar configuration, found that sensor-based wind speed measurements achieved good accuracy for hovering and low-speed vertical movement, however they were not shown to be suitable for high-speed (> 2 m s⁻¹) vertical flight patterns. Alternatively, an attitude-based approach showed good accuracy also for high-speed vertical flight patterns. In both cases, platform configuration significantly affected wind speed readings, highlighting the necessity of configuration specific calibrations. Here, the intercepts and slopes were used for correcting our measurement data.

As presented in and , a reduction in uncertainty for wind speed and direction can be achieved as averaging time increases. In our case, for both wind speed and direction, increasing the averaging time to 10 s does not significantly reduce uncertainties in relation to the trade-off with flight time and desired spatial resolution. Thus, we use a 5 s averaging interval for our following measurements.

T and RH are also important parameters in meteorological sounding as they are commonly used to discern atmospheric stratification . T and RH measured during the same testing period exhibited a very good agreement (Figs. S3, S4) (RMSE in the order of 0.1 °C and 0.1 % RH). Indeed, T and RH sensors infer less uncertainty when mounted on a drone compared to an anemometer. evaluated measurement errors of T and water vapor mixing ratio for different unshielded sensor mounting positions on a drone, reporting the largest disagreement with a spread of 0.91 K ≤ T ≤ 1.88 K and -0.22 g kg⁻¹ ≤ qv ≤ 0.66 g kg⁻¹ at the sensor placement associated with the highest uncertainty. However, by creating some distance between the sensors and the drone battery as well as avionics, uncertainty could be reduced further. More critical, however, is response time of sensors in relation to movement speed of the drone, which determines the sensing capability of a drone platform within the desired flight pattern.

The vertical profiles of potential T, specific humidity, wind speed, and wind direction observed during multiple flights over a day-to-night transition period are shown in Fig. . Potential T was used here as it is a common parameter used to estimate boundary layer height . Potential T represents the T an air parcel would have when adiabatically brought to a reference pressure (e.g 1000 hPa). As opposed to tracking T, potential T is used to evaluate the atmosphere's actual vertical T gradient, excluding any diabatic change in the probed air. The observed mean potential T per profile dropped significantly from 274 K (min = 274.3, max = 274.8) at 19:30 (all times are given in local time, UTC+1) to 269 K (min = 268.9, max = 269.9) at 02:30 (Fig. a) and correspondingly water content in the air increased from 2.5 g kg⁻¹ (min = 2.4, max = 2.4) at 19:30 to 3.0 g kg⁻¹ (min = 2.9, max = 3.1) at 02:30 (Fig. b). Variability in profiles of potential T is low (min and max within 0.5 K) between 19:30 and 20:40, however, small but detectable changes at 100 m at 02:30 are observed (min = 268.9, max = 269.9 K) (Fig. a). Which coincides with an enhanced variability in humidity at the same height and time. It is worth noting that interference from the surface (first measurement point represents average and SD from 0 to 5 m a.g.l.) is not significant for all parameters, higher variability can be seen mainly for wind speed and direction, however only in profiles 19:30, 20:40, and 23:30. Wind speed exhibits notable variation between 90 and 110 m a.g.l. at 20:40 which also correlates to relatively large variation of humidity at the same height and time (Fig. b & c). Wind direction exhibits comparatively greater variability at 02:30 specifically between 20 and 50 m a.g.l. (Fig. d). The data collected during descending flights can be potentially poor due to the drone travelling into its own generated downwash, a known issue for drone-based sensor measurements . Thus, only the vertical profiles obtained during ascent were used. In the case that the nocturnal lower boundary layer is more clearly stratified, the first 120 m a.g.l. of the troposphere can exhibit much stronger dynamics (as shown in Fig. ) than observed in Fig. .

In Fig. , we present the impacts of drone hovering on sampling with two comparisons: drone inactive (no operations) versus ground-based measurements, and drone hovering versus ground-based measurements. For both conditions, the drone-based and ground-based filter samples show strong agreement with Pearson's R 0.99 and 0.97 for hovering and inactive (i.e. grounded) conditions, respectively. No apparent difference is observed between measurements performed during hovering conditions and while the drone was inactive and grounded, highlighting the negligible rotor-induced perturbations on the collection efficiency and the consistent analytical reproducibility even relatively clean environments. We also notice the uncertainty of the measurement especially for low signal species. While the average differences between drone and ground-based measurements ((|Drone-Ground|)/Ground) for all compounds are 13.4 % and 14.1 %, for grounded and hovering conditions respectively, the compounds with strong signal have much less relative difference compared to the compounds with weak signals, as demonstrated in Fig. S2. This highlights the necessity of enough particle loading for reliable measurements.

To validate the comparability of drone and ground-based sampling in extracting volatility information, the thermogram shape (the amount of a certain chemical species being desorbed as T increase) and Tmax⁡ of individual compounds from the drone and ground-based filters were compared. Tmax⁡ from the drone-based filter vs ground-based filter shows a good correlation (R=0.8) with a slope of 0.98 and an intercept of 3 °C. In Fig.  four major organic and two major inorganic peaks are shown, area overlap between drone and ground-based thermogram shape is consistently over 92 %, with the strongest agreement seen for HNO₃ (agreement > 99 %). This emphasizes the comparability in volatility information extracted from the drone and ground-based filter collections.

To demonstrate the drone's ability to detect nocturnal boundary layer dynamics and sample particles aloft, a representative measurement case under stable nocturnal conditions is presented. 5 vertical profiles of wind speed, wind direction, potential T and specific humidity were performed in parallel to ground and drone-based aerosol particle sampling. The vertical profiles revealed a shallow boundary layer as determined by a strong surface-based T inversion (Fig. ). The potential T increased by approximately 5 K before returning to near stable lapse rate conditions. This indicates stratification of the airmasses based on known boundary layer dynamics .

In contrast to a non-stratified mixed layer, where airmasses transport atmospheric constituents via convective mixing, very little vertical mixing will occur between layers separated by an inversion layer as the coldest air (highest density) is found at the surface. Hence, sharp differences in particle loading have been observed in relation to T inversion events in different environments and topologies . Here, we observed a significant effect on particle concentration and composition (Fig. ) due to airmass stratification following the formation of a surface-based inversion layer. The total organic signal is a factor of 3 higher at the ground level compared to aloft. This aligns well with previous studies showing a sharp decline in particulate matter at the height of an inversion layer . Dominant organic species, e.g. C₅H₈O₄, C₉H₁₅NO₆ and C₁₀H₁₇NO₇ are an order of magnitude higher at the ground level. The high signals of nitrated species are consistent with the expectation that nearby NO_x emissions during the night are accumulated close to the surface under stable nocturnal boundary-layer conditions , which lead to formation of organic nitrates from both freshly emitted VOCs and those remaining from daytime emissions . The concentration of C₆H₁₀O₅ (mainly levoglucosan ), a common biomass burning tracer, is also much higher in the ground-based sample. Other potential biomass burning species such as nitrophenol (C₆H₅NO₃), methyl nitrophenol (C₇H₇NO₃), nitrocatechol (C₆H₅NO₄), and methyl nitrocatechol (C₇H₇NO₄) are detected in the ground sample at similarly higher concentrations compared to aloft, although not as prominently as levoglucosan. This indicates a potential effect on the dispersion of biomass burning aerosol from the restriction of airmass vertical transport due to the nighttime T inversion. Furthermore, notable inorganic species such as nitric and sulfuric acid (HNO₃, H₂SO₄) are also stratified, with HNO₃ being almost 50 % more abundant at the ground level, but H₂SO₄ four times more abundant at 120 m, indicating different source or sink processes.

While nitric acid accumulates at a ground level, sulfuric acid emissions may have different sources, e.g from taller anthropogenic emissions sources and plumes (i.e. emission stacks or scrubber towers). Moreover, comparing the summed thermogram for ground and aloft measurements reveal a difference in the overall Tmax⁡ by 15 °C from 75 °C (ground) to 90 °C (aloft). Indicating a lower volatility in aerosol sampled aloft compared at the ground level. More comprehensive results will be presented in our subsequent publication on nocturnal boundary layer chemistry.

We developed a novel drone platform to sample particulate matter for molecular-level chemical characterization using a FIGAERO-CIMS as well as perform meteorological sensing within the PBL. Following previous work on offline filter collection for FIGAERO-CIMS analysis and many applications in field campaigns this platform greatly improves the sampling reach of the instrument. Here, we demonstrated its capability in filter collection and meteorological sounding by validating drone mounted sensors against stationary measurement tower readings, as well as intercomparing ground-based and drone-based filter sampling system.

All sensors (T, RH, and wind speed and direction) displayed good agreement with stationary measurements. Different averaging times, which also relates to the spatial resolution of measurements during in-flight measurements, have been previously documented for similar setups . Here, only slight improvements are gained when increasing averaging times from 5 to 10 s. Vertical profiles conducted during the day-to-night transition further demonstrated their good performance, successfully capturing atmospheric changes.

Filter collection efficiency and comparability for the drone and ground-based systems were evaluated by collocated sampling. This comparison for both hovering and inactive conditions showed a very good comparability with negligible effects of drone operation on collection efficiency, highlighting the applicability of drone platforms to perform filter sampling. The comparison of thermograms, i.e volatility information, from drone and ground-based measurements, showed excellent agreement across identified compounds.

Finally, the deployment of the drone platform during nighttime demonstrates its capability to capture vertical gradients in both meteorological parameters and aerosol molecular composition. A surface-based T inversion was detected, where potential T increased by 5 K in the first 120 m a.g.l. Filter samples collected at the ground and drone level showed significant differences in composition and signal intensity. Total organic signal was higher at the ground level, differing by a factor of 3. Dominant organic species C₆H₁₀O₅ (likely levoglucosan), C₉H₁₅NO₆ and C₁₀H₁₇NO₇ (nitrated monoterpenes) and C₅H₈O₄ (oxygenated isoprene product) differed by an order of magnitude. Further, inorganic species such as HNO₃ and H₂SO₄ are also seemingly stratified, where HNO₃ is more abundant at the ground level, and H₂SO₄ is more abundant aloft.

The drone platform presented here offers greatly improved spatial coverage for aerosol particle sampling as compared to the conventional online and offline sampling system. It offers the possibility to sample in otherwise inaccesible locations and altitudes at a low cost and high mobility, improving direct characterization of the vertical variability in aerosol chemical composition. Naturally, the platform is not without its limitations, payload capacity and flight time are constrained by technical progress and filter collection can be challenging in very clean environments. Nevertheless, the platform presented here provides novel insights into drone operation and enabling flexible, extended filter collection, which can be widely applicable and adaptable to other chemical analysis techniques.