The aircraft platform used to detect aerosols and cloud droplets consists of an airplane (King-Air 350) and several onboard instruments manufactured by Droplet Measurement Technology of the United States. Their probes and functions are listed in Table 1. The Cloud Droplet Probe (CDP) can detect 2–50 µm liquid cloud droplets or particles (30 bins). The Cloud Imaging Probe (CIP) can detect ice crystals and big cloud droplets ranging from 25 to 1550 µm (25 bins). The Cloud Condensation Nuclei Counter can measure CCN number concentrations at various supersaturation levels. The Passive Cavity Aerosol Spectrometer Probe (PCASP) can detect particles ranging from 0.1 to 3 µm (30 bins). In addition, the Aircraft-Integrated Meteorological Measurement System was employed to collect ambient temperature, relative humidity, wind speed, wind direction, and other meteorological factors.

Figure 1 shows a 3-D full flight track and its vertical view over southern Anhui on 1 November. The airplane took off from the Jiuhuashan airport (117.68°E, 30.74°N) at 14:24 local time (henceforth, LT)and then flew horizontally at 2000 m height towards Huangshan between 14:34 LT and 14:46 LT. After that, the airplane spiraled up to a maximum altitude of 5679 m, during which it passed through four cloud layers, a low layer at 970–1000 m, a middle layer at 2240–2500 m, a mid-high layer at 3400–4050 m, and a high layer at 4350–5400 m (Figure 1A). Later, the airplane started to conduct artificial seeding for precipitation enhancement at 15:06 LT, with an approximately horizontal “8”-shape cross closed track (A-B-C-D-B1-E-A) around 5000 m (Figure 1B). The artificial seeding lased two circles with nearly the same trajectory. According to records, freezing occurred on the airplane wings above 5300 m at 15:13 LT, and heavy icing appeared at 15:19 LT, indicating abundant supercooled water within high clouds. During the flight, air temperatures were −4 ∼ −6°C within seeded clouds. The seeding operation released 40 AgI strips (seeding agent), about 1440 g in total, with a burning time of 16 min for each one. The seeding experiment finished at 16:07 LT, and subsequently, the airplane returned to the airport.

The cloud properties, such as cloud top height, brightness temperature, cloud type, and cloud phase, were obtained from Himawari-8 (https://www.eorc.jaxa.jp) and FY-4 (http://data.nsmc.org.cn) satellites. The quality of precipitation datasets have been greatly improved in Integrated Multisatellite Retrievals for GPM (IMERG) which provides seamless precipitation estimates at 0.1 × 0.1 grid every half hour (https://gpm.nasa.gov/data). Reanalysis data were from the National Centers for Environmental Prediction (https://psl.noaa.gov/data). Aerosol optical depth (AOD) at 550 nm and aerosol classification data were from the version 2 Modern-Era Retrospective Analysis for Research and Applications (MERRA-2, https://disc.gsfc.nasa.gov). MERRA-2 provides five types of aerosols at 72 terrain-following hybrid σ–p model layers, including dust and sea salt in five bins, sulfate, organic carbon, and black carbon (Buchard et al., 2017; Randles et al., 2017). Since this study focuses on submicron aerosols, here we only calculate the first two bins of dust and sea salt. The mass concentrations of PM_2.5 were obtained from the China National Environmental Monitoring Center (https://www.cnemc.cn).

Based on particle size, the droplets less than 50 μm measured by CDP are defined as small cloud droplets, and their number concentrations and effective diameters are recorded as N _c (cm⁻³) and D _c (μm). The droplets larger than 50 μm measured by CIP are defined as big cloud droplets, and their number concentrations and effective diameters are recorded as N _cip (m⁻³) and D _cip (μm). The droplets larger than 100 μm can be treated as ice crystals (Dong et al., 2021). Due to the low accuracy of the first bin (0.1–0.11 μm), it was excluded from the total measurements of PCASP, so aerosols mainly include accumulation (0.11–1 μm) and coarse (1–3 μm) particles. In clouds, aerosol detection by PCASP reveals more uncertainty compared to outside clouds due to supersaturation conditions, where part of particles can act as CCN and convert into cloud droplets (Kleinman et al., 2012; Yang et al., 2019). To avoid errors, we screened the data observed inside clouds in the following analysis of aerosol vertical distributions. The profiles of vertical data were averaged at an interval of 50 m.

Previous studies have proposed many cloud mask algorithms to detect clouds, such as N _c > 10 cm⁻³ (Rangno and Hobbs, 2005), LWC >0.01 g m⁻³ (Gultepe et al., 1996), and N _c > 0.1 cm⁻³ and LWC >0.0005 g m⁻³ (Gultepe and Isaac, 2004). To eliminate the influence of large aerosol particles on N _c measurements, we used the method of Zhang et al. (2011) to identify cloud appearance, namely thresholds of N _c > 10 cm⁻³ and LWC >0.001 g m⁻³.

Precipitation occurred on 1 November over southern Anhui. One cold vortex originating from northeast areas moved southward and strengthened the transport of cold air and water vapor, meanwhile a low-level convergence enhanced in east China. Wind shear at 850 hPa caused strong upward air motion at this level. The targeted area is located in the westerlies, controlled by winds primarily from west and southwest at 500 hPa with wind speeds about 20 ms⁻¹, and the relative humidity was above 90% (Figure 2). Affected by the wind direction, the seeded clouds moved eastward.

Figure 3 shows the spatial distribution of PM_2.5 concentrations and AOD on 1 November. Ground-based observations indicated that Henan and northern Anhui were slightly polluted with PM_2.5 of 75–115 μg m⁻³. A possible reason is that air pollution was mainly concentrated in metropolitan areas and heavy industry areas (Miao et al., 2022). The pollutants from these regions can be transported eastward to downstream cities such as Shanghai. High AOD values ranging from 1.2 to 1.6 were found in one belt from southwest to northeast, corresponding to the airflow field at 500 hPa.

Figure 4 presents cloud macro properties at 15:00 LT on 1 November. According to images of Himawari-8, the cloud cluster covered the belt with high AOD (Figure 3B). The clouds over southern Anhui were dominated by altostratus and nimbostratus (Figure 4A) with cloud top height about 6500 m (Figure 4B). The brightness temperature was −6 ∼ −12°C (Figure 4C). The corresponding cloud phase products of the FY-4 satellite implied the existence of abundant supercooled water in clouds.

Spatial-temporal variations of aerosol composition, size, and concentrations can be influenced by emissions, atmospheric advection and diffusion, conversion of gaseous precursors, and aging processes (Quan et al., 2015; Zhang et al., 2015). Figures 5A–E shows vertical distributions of aerosol particles. Aerosols exhibited vertical stratification and heterogeneity. Large amounts of aerosols were constrained below 1050 m, near the top of the boundary layer, indicating that the upward transmission of aerosols to the free troposphere was limited (Figure 5B). In general, aerosol number concentration (N _a ) decreased with increasing altitude below the bottom of the low-level clouds (1050 m), while aerosol effective diameter (D _a ) keeped stablewith an average of 0.12 μm near the surface (Figure 5C). Since aerosols in the lower atmosphere mainly originate from local emissions, especially under unfavorable dilution conditions, their properties are complex and greatly affected by anthropogenic activities. Sulfate and organic aerosols made up a large fraction of aerosol mass in the lower atmosphere (Figure 6), where aerosols mainly consist of primary and secondary aerosols generated through gas-to-particle conversion (Jimenez et al., 2009; Li et al., 2015; Li et al., 2018). D _a showed an evident maximum between 2500 and 3000 m with an maximum of 1.72 μm (Figure 5C). However, above 3000 m, dust took a higer proportion, D _a was smaller than 2500–3000 m. The possible reason is particle hygroscopicity under high ambient humidity, where an obviously positive correlation (r = 0.58) exists between D _a and supersaturation rate (Figure 7). At other cloudless atmospheric layers, N _a declined slightly with altitude, while D _a remained nearly unchanged. Figure 8 shows the mean aerosol size spectrum observed in five cloudless atmospheric layers. Clearly, fine particles, especially particles with sizes of less than 0.15 μm, predominated at all altitudes. The peak of the unimodal spectrum centers at 0.12 μm. Accumulation mode aerosols account for up to 98% of total number concentrations. Such a result is similar to Hao et al. (2017), in which they found that accumulation mode aerosols accounted for more than 95% during August 2014 over Anhui Province.

Aerosols from several large-scale industrial districts in southern Anhui have been found to have a high hygroscopic growth ability (He et al., 2016). Figure 5D shows vertical distributions of CCN concentrations at a 0.2% supersaturation level. CCN concentrations decreased with altitude from the ground until 4000 m and increased sharply at higher levels. The ratio of CCN to N _a represents the activation ability of aerosols to become CCN under certain supersaturation conditions. Note that due to the detection limit of instruments, the absence of particles with sizes less than 0.1 μm or greater than 3 μm may result in somewhat underestimation for N _a and CCN/N _a exceeding 1.0 at various altitudes. Aerosol CCN activation is known to be closely related to aerosol size, composition, mixing state, and ambient humidity (McFiggans et al., 2006; Bauer and Menon, 2012; Farmer et al., 2015; Kanji et al., 2017). It is also known that aerosol chemical composition and the initial periods of activation play important roles in aerosol CCN activation under low supersaturation conditions, particularly for fine particles with low-hygroscopicity, such as primary organic aerosol (Zhang et al., 2012). However, aerosol CCN activation depends on aerosol size more under high supersaturation ratio due to strong hygroscopicity (Figure 7).

In terms of CCN/N _a and D _a , the cloudless atmosphere can be divided into three layers below 5300 m. Below 2200 m, CCN/N _a increased with altitude and reached the maximum at 2200 m (Figure 5E). CCN/N _a was lowest within the boundary layer (Figure 5E), where aerosols were mainly composed of sulfate and hydrophilic organic carbon with D _a less than 0.12 μm (Figure 6). Moreover, hydrophobic organic aerosols (i.e., black carbon and soot) from local emissions are difficult to activate into CCN (Zhang et al., 2017; Yu et al., 2022). The CCN ability of low-hygroscopic organic aerosols can be significantly enhanced under pollution due to their mixing with local and regional pollutants (Hu et al., 2020). Between 2200 and 3000 m, CCN/N _a was stable. Above 3000 m, CCN began to increase and CCN/N _a was relatively higher than the boundary layer due to a higher proportion of dust transported from upstream regions (Figure 6). Li and Shao. (2009) reported that aerosol aging and secondary material generation occur in long-distance transportation of air mass. Dust hygroscopicity increases after its aging, and aged dust aerosol become activated to be CCN in the atmosphere (Tang et al., 2015).

The zero level (0°C) of ambient temperature appeared at 4650 m (Figure 5A). There were four layers of clouds located at different heights in the atmosphere (Figure 1A, Figure 5). Specifically, the maximum N _c was 125 cm⁻³ at 2284 m in the mid-level clouds (Figure 5F), and the maximum D _c was 34.25 μm at 4777 m in the high-level clouds (Figure 5H). Overall, inside each cloud layer, N _c and D _c increased with altitude and reached the maximum in the middle, and then decreased with altitude until the cloud top. Such variability is related to the entrainment of drier air from the environment that induces fewer cloud droplets by promoting evaporation (Freud et al., 2011).

The low-level clouds at the top of the boundary layer were very thin, with a mean N _c of 25.21 cm⁻³ (Figure 5F). At this level, a large number of fine particles uplifted from the boundary layer are difficult to grow into large cloud droplets due to the competition for limited water vapor (Yuan et al., 2008; Grandey and Stier, 2010), and therefore, cloud droplets sizes were small, with a maximum D _c of 12 μm (Figure 5H). The mid-level cloud layer was also thin, but the maximum N _c and the minimum D _c appeared here approximately 107.7 cm⁻³ and 4.03 μm, respectively. According to Swietlicki et al. (2008), hygroscopic particles are mainly composed of inorganic salts like sulfates. The average CCN/N _a of aerosols was about 0.78 at 200 m below the mid-level cloud base (Figure 5E). Sulfate accounted for 69.8% of the aerosol mass (Figure 6), indicating that the aerosol particles here have a high ability of hygroscopic growth to activate into cloud droplets when entrained into the cloud, resulting in more cloud droplets with smaller size (Jones et al., 1994; Ishizaka and Adhikari, 2003). The mean N _c and D _c of the mid-high level clouds were 24.73 cm⁻³ and 16.46 μm, respectively, while they were 12.54 cm⁻³ and 16.65 μm in the high-level clouds. Although the mid-high and high-level clouds had relatively low cloud droplet loading, their D _c was relatively larger due to sufficient LWC and lower temperature.

Both cloud droplet size spectra in the low- and mid-level clouds showed a single peak at 5 μm, and no droplets greater than 12 μm appeared (Figure 5J). By contrast, in the mid-high layer cloud, the spectra showed a bimodal pattern, with the first peak at 8 μm and the second peak at 15 μm. Moreover, a trimodal pattern with peaks at 7 μm, 12 μm, and 17 μm appeared in the high-level clouds. Relative dispersion rate (ε), defined as the ratio of the standard deviation of cloud droplet size spectrum and average radius of cloud droplets, is used to illustrate the relative spectral width of cloud droplet spectrum. It is usually linked with the physical and chemical properties of aerosols, related activation process and hygroscopic growth (Ma et al., 2010; Wang et al., 2019), the development stage of clouds, atmospheric temperature, humidity, entrainment process, and turbulence (Lu et al., 2013). A positive correlation was found between the relative dispersion rate and D _c (Figures 5H,I, r = 0.78). In addition to the coalescence process (Liu and Daum, 2002; Pandithurai et al., 2012), the appearance of ice crystals with large sizes also increases cloud droplet inhomogeneity which leads to a broadening cloud droplet size spectrum in higher cloud layers.

To examine cloud evolution during the artificial seeding experiment, we compared the cloud microphysical properties detected by CDP and CIP (Figure 9). The cloud seeding began at 15:06 LT. Before that, cloud droplets smaller than 25 μm prevailed, accounting for about 97%, and cloud droplets greater than 50 μm hardly existed. The size spectra of cloud droplets showed a trimodal pattern, with the first peak at 7 μm, the second at 12 μm, and the third at 17 μm, indicating sufficient supercooled water in clouds. Moreover, there was a negative relation between D _c and N _c , indicating that the effect of strong coalescence caused some small cloud droplets to convert into large droplets and ice crystals. The man-made ice nuclei seeded into the clouds grew through the Wegener–Bergeron–Findeisen mechanism (Wegener, 1911; Bergeron, 1935; Findeisen, 1938). Supercooled water droplets were converted into ice crystals, resulting in the evaporation and dissipation of small cloud droplets and the formation of large cloud droplets or ice crystals (Figure 10).

Considering the horizontal movement of air mass near the flight track line, seeding agent diffusion, and cloud spatial inhomogeneity, we selected the A–B and B1–E sub-tracks (Figure 1B) to compare changes in clouds before and after artificial seeding. Before seeding, mean N _c and D _c were 20.87 cm⁻³ and 15.98 μm, respectively, and CIP almost detected nothing. After seeding, the mean N _c reduced to 16.62 cm⁻³ in the A-B sub-track and 6.36 cm⁻³ in the B1-E sub-track, while the mean D _c was raised to 23.33 μm and 38.90 μm in these stages. Meanwhile, the mean N _cip increased slightly to 20 L⁻¹ and 70 L⁻¹ in two sub-tracks. Especially, cloud droplets bigger than 25 μm increased significantly from 3% to 16% in the A-B sub-track and to 56% in the B1-E sub-track, indicating a broader cloud droplet size spectrum because of condensation growth. The initiation of precipitation is often considered to occur when R _c exceeds 12–14 μm (Rosenfeld et al., 2012; Bera et al., 2016; Braga et al., 2017), corresponding to Dc of 24–28 μm (Sheng et al., 2022). During this experiment, N _c decreased while D _c increased, owing to the AgI agent that worked as artificial ice nuclei to enhance the condensation growth of cloud droplets. Ice crystals continued to grow and formed ice crystal coagulation through condensation, riming, and collision, thus widening the cloud droplet size spectrum (Rosenfeld and Bell, 2011; Chen et al., 2017; Lerach and Cotton, 2018). In this conversion process, latent heat was released due to seeding-induced glaciation to strengthen updraft in clouds, and thereby some cloud droplets and ice crystals were brought to higher levels.

Satellites provide products with high spatio-temporal resolutions to examine cloud and precipitation changes in real situation (Rosenfeld et al., 2019; Yue et al., 2019). Reananlysis data can be regard as the case of cloud and precipitation not influenced by cloud seeding. We calculate the 1-h air-moving influence area affected by the seeding agent through advection and diffusing with wind speed about 20 m s⁻¹ above 5000 m. Figure 11 shows the cloud droplet effective radius observed by the Himawari-8 satellite at 15:00 LT (before seeding) and at 16:00 LT (after seeding). Before seeding, cloud droplets smaller than 20 μm predominated, and droplets greater than that only accounted for 5%. After seeding, cloud droplets larger than 20 μm increased notably up to 15% over downstream regions. In Figure 12, the spatial average of hourly precipitation over air-moving influence area is stable around 0.2 mm based on reanalysis data. However, comparing with satellite products, an obviously increment of precipitation shows after cloud seeding (15:00 LT) with maximum up to 1.05 mm. On the whole, the artificial precipitation experiment enhanced rainfall, reduced cloud cover, and contributed to the removal of pollutants from the atmosphere through wet scavenging, which was beneficial to the air quality in the downstream regions (e.g., Shanghai). Previous studies have shown that precipitation has a significant wet removal effect on aerosols of different sizes, especially fine particles (Zhao et al., 2015; Guo et al., 2016). Lu et al. (2019) analyzed the PM_2.5 removal effect by precipitation with different intensities and duration, raindrops spectrum, and wind speeds and found that strong precipitations often exert higher removal rates on aerosols. Since Huangshan, Xuancheng, and Nanjing are located downstream of the seeding operation area (Figure 1B), air quality in these cities was improved by the experiment. PM_2.5 decreased since 15:00 LT in Huangshan and then followed with a 1 hour interval in Xuancheng and Nanjing, respectively (Figure 13A), corresponding to the air-moving influence area. Comparing with PM_2.5 before cloud seeding, it decreased by 23%, 18% and 26% at most in Huangshan, Xuancheng, and Nanjing, respectively. However, PM_2.5 of Fuyang, Huaian, and Luan located in northern Anhui, out of the seeding operation area, was at high levels and worsening (Figure 13B). This result demonstrated that the artificial seeding experiment could reduce pollutants by wet scavenging and mitigate pollution in downstream regions efficiently. It is worth noting that how to more quantitatively evaluate the contribution of artificial seeding precipitation and natural rainfall to the removal of pollutants is an issue that warrants further studies.