Observation data for this study were collected from 1 June 2021 to 30 May 2022, during comprehensive atmospheric environmental observations conducted at the Fujian Provincial Environmental Monitoring Center Station (26.11° N, 119.30° E, altitude 65 m) and the Fuzhou Meteorological Bureau Station (26.05° N, 119.26° E, altitude 18 m). Both stations are located within Fuzhou's urban area, approximately 8 km apart horizontally. The Fujian Provincial Environmental Monitoring Center Station is situated in Gulou District, the central urban area of Fuzhou, surrounded primarily by commercial, residential, and transportation land, representing areas heavily influenced by intense human activities. The Fuzhou Meteorological Bureau Station is located in Cangshan District, southern Fuzhou, approximately 1.5 km east of the Min River. Fuzhou is situated at the Min River estuary and along the East China Sea coast, characterized mainly by plains (average altitude 10–30 m) and a typical East Asian monsoon climate, significantly influenced by sea-land breeze circulation and marine air masses (Hu et al., 2024). Given the short distance (∼8 km) and the regional background homogeneity, the local environments of the two sites are not expected to be drastically different, although each is influenced by its specific surroundings (urban vs. riverside). A comparison of air pollutant concentrations between the two sites using nearby national monitoring stations (Table S2 in the Supplement).

A CCN counter (CCN-100; DMT, USA) equipped with a continuous flow of 500 cm3min-1 and a thermal gradient was used to measure CCN concentrations at five SS levels. To maintain counting accuracy, the instrument was regularly calibrated for T gradient, flow rate, pressure, SS, and the optical particle counter (OPC) using standard ammonium sulfate according to the method by Rose et al. (2008). Additionally, zero-point determination was performed before and after each observation to minimize instrumental error. During observations, the measurement interval for each SS level was 10 min, and a few minutes were required to stabilize after switching SS levels. Therefore, CCN data collected before reaching stable SS were excluded from subsequent analysis. The typical measurement uncertainty of the CCN-100 is approximately ±10 % in Table S1.

An online organic carbon/elemental carbon analyzer (Sunset Laboratory semicontinuous OC/EC analyzer, Model-4, Sunset Laboratory Inc., USA) was used to determine organic carbon (OC) and elemental carbon (EC) content in atmospheric particulate matter samples. The instrument provides hourly averaged concentrations. Details of instrument operation can be found in Chang et al. (2017). Based on replicate analyses reported by Zhang et al. (2021), the within-model measurement uncertainties (relative standard deviation) for the Sunset Model-4 are ±3.6 % for OC and ±6.8 % for EC.

A Wide-Range Particle Spectrometer (WPS-1000, MSP) measured aerosol number size distributions (10–350 nm) with a time resolution of 6 min across 96 channels. To align with the 1 h meteorological data, the 6 min distributions were arithmetically averaged to hourly particle number size distributions. Instrument details and principles are described in Wang et al. (2014). The typical uncertainty for particle size distribution measurements in this size range is approximately ±10 % in Table S1.

Black carbon (BC) mass concentration was measured using an Aethalometer (AE-33, Magee Scientific) with a time resolution of 1 h. BC data from the 880 nm wavelength (channel 6) were used (Kirchstetter et al., 2004). The measurement uncertainty is approximately ±10 % (Table S1).

An online particle chromatograph (MARGA ADI-2080) continuously monitored mass concentrations of soluble aerosol ionic components (SO42-, NO3-, NH4+, Na+, K+, Ca2+, Cl-) and trace gases (NH3, HNO2, HNO3, HCl, SO2). Sampling, operation, and internal calibration methods followed Du et al. (2010). The instrument provides hourly averaged concentrations. According to an independent verification study (Battelle, 2009), the measurement precision (median absolute relative percent difference between duplicate units) ranges from 5 % for SO2 to 20 % for NH4+, with data completeness >90 % for all major ions.

Meteorological data (including wind speed (WS), wind direction (WD), temperature (T), relative humidity (RH), and precipitation) with a time resolution of 1 h were obtained from the Fuzhou Olympic Sports Center Meteorological Station. Data on conventional air pollutants (O3, CO, NO2, PM_2.5, and PM₁₀) were sourced from the China National Environmental Monitoring Centre's real-time urban air quality release platform (https://quotsoft.net/air/, last access: 26 May 2026).

The growth rate (GR) of new particles was calculated following (Kulmala et al., 2012): 1GR=ΔDmΔt where Dm is the median diameter of the nucleation mode particles, obtained by fitting a log-normal distribution to the particle number size distribution. Statistics of the fitting results (see the Supplement) demonstrate that the log-normal distribution represents the observed particle size distributions well, with the majority of fits yielding high coefficients of determination (R2>0.85); only fits with R2>0.7 were used to avoid propagating poor-fitting uncertainties. GR describes how rapidly particles grow from the nucleation size to larger sizes. It is later used to estimate the formation rate of new particles (Eq. 4) and condensable vapor concentration (Eq. 5) and to evaluate the competition between condensation and coagulation during NPF events.

The condensation sink (CS) reflects the rate at which condensable vapor molecules condense onto the surface of pre-existing atmospheric particles and was calculated as follows (Kulmala et al., 2012): 2CS=4πD∑iβM,i⋅Di⋅Ni where D is the diffusion coefficient of the vapor (typically assumed to be sulfuric acid), Ni is the number concentration of particles in a given size bin, and βM is a correction factor.

The coagulation sink (CoagS) reflects the ability and rate of pre-existing atmospheric particles to remove newly formed nucleation particles via coagulation. For particles of size i, the coagulation sink can be expressed as: 3CoagSi=∑jKijNj where Nj is the number concentration of particles in size bin j, and Kij is the Brownian coagulation coefficient between particles of size j and i.

The formation rate (FR) of new particles was calculated following Kulmala et al. (2012): 4FR=dNnucdt+CoagSnuc⋅Nnuc+GRΔdp⋅Nnuc+Slosses where Nnuc is the number concentration of nucleation-mode particles. Following the definition by Kulmala et al. (2012), the nucleation-mode size range in this study was also limited to below 25 nm. CoagSnuc⋅Nnuc is the flux of particles lost due to coagulation with pre-existing particles, where CoagSnuc is the coagulation sink for nucleation-mode particles. GR/Δdp represents the flux of particles growing out of the nucleation size range (exceeding 25 nm), which is generally negligible under typical atmospheric conditions (Dal Maso et al., 2005). FR is later used to compare NPF event intensity under different meteorological and chemical conditions, and to identify periods with active nucleation. The additional loss term Slosses (e.g., dilution due to boundary layer growth, wall losses) was negligible under our field conditions (Dal Maso et al., 2005). For regional NPF events, transport losses can also be ignored. FR is a direct measure of NPF intensity and is later compared across different meteorological and chemical conditions.

Condensable vapor concentration (C) and source rate (Q). Assuming that particle growth is dominated by condensation of a low-volatility vapor (typically sulfuric acid), the vapor concentration can be estimated from the observed growth rate (Dal Maso et al., 2005; Kulmala et al., 2012): 5C=A⋅ddpdt where dp is the particle diameter, and A is a constant, which has the value 1.37×10-7 hcm-3nm-1 for a vapor with molecular properties of sulfuric acid (Dal Maso et al., 2005). This provides an upper-limit estimate of the condensable vapor concentration, as it assumes growth is solely due to condensation of the vapor and neglects contributions from coagulation.

The CS (Eq. 2) quantifies the rate at which this vapor is removed by pre-existing particles. Under steady-state conditions (dC/dt=0), the vapor source rate Q can be derived as (Dal Maso et al., 2005): 6Q=CS⋅C This source rate represents the net production of condensable vapor needed to maintain the observed growth and is later compared with precursor gas concentrations (e.g., SO2) to infer the chemical pathways driving NPF.

Hygroscopicity parameter (κinorg) for inorganic species. Due to the lack of organic composition measurements, we estimated the hygroscopicity of the inorganic fraction only. The measured water-soluble ions (SO42-, NO3-, NH4+, Cl-) were converted to mass concentrations of inorganic salts using the ion-pairing scheme described in Gysel et al. (2007) and Kuang et al. (2020). The following salts and their κ values (Kuang et al., 2020, Table S2) were considered: (NH4)2SO4 (κ=0.48), NH4NO3 (κ=0.58), NH4HSO4 (κ=0.56), and NH4Cl (κ=0.93). The volume fraction of each salt was calculated using its density (also from Kuang et al., 2020). The overall inorganic hygroscopicity parameter κ inorg was then obtained by volume-weighted mixing (Petters and Kreidenweis, 2007): 7κinorg=∑iεiκi where κi and εi represent the hygroscopicity parameter and volume fraction of component i in the mixture, respectively, and i denotes the number of components. This κinorg represents the hygroscopicity of the inorganic aerosol components and is used as an upper limit estimate for the total particle hygroscopicity, as organic matter (typically less hygroscopic) was not included.

The concentrations of secondary organic carbon (SOC) and primary organic carbon (POC) were estimated following (Wu and Yu, 2016): 8POC=(OC/EC)min⋅EC9SOC=OCtotal-(OC/EC)pri⋅EC where OCtotal is the measured OC, (OC/EC)min is the minimum (OC/EC) ratio during the observation period, POC is primary organic carbon, and SOC is secondary organic carbon.

The enhancement effect on cloud condensation nuclei number concentration (E_NCCN) was defined as the ratio of CCN number concentration after the NPF event to that before the event (Ren et al., 2021): 10E_NCCN=NCCN, after/NCCN, prior where NCCN, after, is the average CCN number concentration during the NPF event (from its start to end), and NCCN, prior, is the average CCN concentration during the 2 h before the event. This factor directly links NPF to potential cloud formation: a value >1 indicates that NPF increases CCN availability.

XG Boost-SHAP framework. To quantitatively evaluate the nonlinear effects of meteorological factors (T, RH) and precursor gases (NH3, SO2) on the particle formation rate (FR), we applied an interpretable machine learning framework combining XG Boost (Extreme Gradient Boosting) with SHAP (SHapley Additive exPlanations). A detailed description of the feature selection, model training, validation, and SHAP interpretation is provided in the Supplement (Sect. S1). The main quantitative thresholds and interaction strengths derived from this analysis are discussed in Sect. 3.4.

NPF events were identified based on criteria from Kulmala et al. (2012): (1) significant increase in nucleation-mode number concentration (Nnuc) (diameter 10–25 nm); (2) formation of a new mode lasting several hours; (3) growth of the newly formed mode over several hours. Additional criteria for NPF identification included: low pre-existing particle number concentration, a clear “banana-shaped” evolution in particle number concentration over time and size, and exclusion of interference from pre-existing particles (especially in urban environments) (Heintzenberg et al., 2007).

In this study, a day was defined as an effective NPF day if the nucleation-mode (10–25 nm) particle number concentration increased continuously for at least 2 h from its initial value to its maximum and showed clear growth to larger sizes (e.g., 12–50 nm) over several hours (Fig. 1). Other days were defined as non-NPF days (Leng et al., 2014). During the one-year observation, a total of 46 NPF events and 319 non-NPF days were identified.

NPF events in Fuzhou exhibited a distinct seasonal preference. As shown in Fig. 2a, spring was the season with the highest NPF frequency (27.17 %), while summer had the lowest (4.35 %). Fall (9.89 %) and winter (8.89 %) showed intermediate to low frequencies. NPF events mainly occurred between 09:00 and 12:00 local time (LT).

NPF events led to significant increases in nucleation-mode (Nnuc) and Aitken-mode (Nait) particle number concentrations (Fig. S1 in the Supplement). Spring showed the highest increase in Nnuc (196.7 %), while fall showed the highest increase in Nait (70.5 %), indicating differences in new particle formation and subsequent growth across seasons.

NPF days typically corresponded to a lower background of PM_2.5 and PM₁₀ (Fig. S3), indicating that NPF occurred in relatively clean atmospheres with a weak condensation sink (CS). It should be noted that the absolute PM_2.5 and PM₁₀ in winter and spring were still relatively high compared to other seasons (Fig. S3), implying that even on NPF days, these seasons experienced higher pollution levels. In spring, summer, and winter, POC on NPF days were lower than on non-NPF days, corroborating this point. Meanwhile, in spring, summer, and fall, SOC was higher on NPF days, suggesting that secondary organic vapors may actively participate in particle formation and growth.

Although spring and winter had heavy background pollution, Fig. S4 shows that the gaseous precursor SO2 concentration on NPF days was significantly higher than on non-NPF days (winter: 0.88 µgm-3 vs. 0.76 µgm-3; spring: 0.64 µgm-3 vs. 0.53 µgm-3). This indicates that in polluted seasons, high gaseous precursors can overcome the inhibitory effect of a high CS and thus trigger nucleation. In contrast, NPF days in summer and Fall exhibited a distinctly clean background, with NH3 and HNO2 significantly lower than on non-NPF days (Fig. S4). In summer, fall, and winter, the concentrations of sulfate (SO42-), nitrate (NO3-), and ammonium (NH4+) on NPF days were significantly lower than on non-NPF days (Fig. S5). This suggests that in these seasons, the lower pre-existing particles and the reduced CS lessen the consumption of gaseous precursors, thereby favouring new particle formation and subsequent growth. However, spring presents a special case in that the SNA (sulfate, nitrate, and ammonium) concentrations on NPF days were comparable to those on non-NPF days (SO42-: 4.35 µgm-3 vs. 4.27 µgm-3). This indicates that spring NPF events are driven by high precursor concentrations. Even when the background particle level is high, the abundant supply of gaseous precursors (Fig. S4) can still overcome the inhibition and trigger NPF events.

According to the hygroscopicity parameter (κinorg) shown in Fig. S6, the κinorg values on NPF days were generally lower than on non-NPF days in all seasons except summer, which is generally attributed to an increased proportion of weakly hygroscopic components in newly formed particles. However, the κinorg on NPF days was significantly higher than on non-NPF days in summer. Despite reduced hygroscopicity, NPF events effectively increased NCCN, which was also combined with higher Cl- concentration on summer NPF days in Fig. S5 (0.59 µgm-3 vs. 0.28 µgm-3). At SS above 0.4 %, NCCN on NPF days was higher than on non-NPF days in all seasons except spring, with the most significant increases observed in winter and fall (Fig. S7).

Figure 3a shows that in spring, particle number concentrations in the 10–20 nm range are generally elevated. NPF typically occurs from 09:00 to 20:00 LT, with the peak concentration of 10–15 nm particles reaching 29 498 cm-3. Concurrently, the concentration of particles >20 nm increases significantly, with some growing beyond 100 nm. The aerosol size distributions on NPF days in summer, fall, and winter all exhibit an NPF process pattern similar to that shown in Fig. 1, typically occurring in 09:00–20:00 LT. During summer NPF events, the maximum particle number concentration reaches approximately 11 410 cm-3, with particles growing up to around 50 nm. In fall and winter, peak particle number concentrations are lower than in summer (10 110 and 5276 cm-3), indicating weaker NPF intensity, and the maximum particle growth can extend up to 100 nm in these seasons.

Spring NPF events had the highest FR (7.13 cm-3s-1) and the highest CS (4.1×10-2 s-1) among all seasons (Table 1). Before NPF events (-4 to 0 h), Nnuc increased from 5425 to 7701 cm-3, peaking at 1 h (8840 cm-3), then gradually declining. This trend was consistent with changes in the FR (Fig. 4a). FR rose sharply to 6.31 cm-3s-1 at the onset of NPF (0 h) and peaked at 7.21 cm-3s-1 at 1 h, demonstrating strong new particle formation capability. However, intense competition under a high CS background significantly suppressed subsequent growth. The growth rate (GR) exhibited large fluctuations, reverting to a negative value at 3 h after an initial peak. Nait increased by only 32 % from 0 h to its peak at 2 h, much lower than in other seasons. Spring NPF events were characterized by strong formation but suppressed growth under high CS.

Despite the low average FR (0.40 cm-3s-1, Table 1), summer exhibited the highest average GR (4.20 nmh-1), and the lowest average CS (1.8×10-2 s-1). In summer NPF events (Fig. 4b), Nnuc peaked at 1 h (3459 cm-3). FR relatively high from 0–2 h (0.97–0.44 cm-3s-1). The most prominent feature of summer was the high growth efficiency (GR) under the lowest CS, with a maximum peak of 11.68 nmh-1 at 2 h. After NPF onset in summer, Nait peaked at 3 h (6461.6 cm-3), with the highest increase of 202.91 %. This indicates that the growth process of new particles in summer NPF events was far stronger than particle formation.

Winter had the lowest average FR (0.23 cm-3s-1) and GR (1.87 nmh-1, Table 1). Winter NPF events were characterized by a low FR and delayed growth under a low condensation sink (CS), as shown in Table 1. In Fig. 4d, the FR peak observed at 0 h (0.43 cm-3s-1) was the lowest among all seasons, indicating weak nucleation. In contrast, the growth rate (GR) displayed a distinct multi-peak pattern, with an initial peak at 2 h (3.20 nmh-1) and subsequent peaks occurring between 5 and 9 h, suggesting that different mechanisms may have driven particle growth at different stages. Correspondingly, Nait reached a maximum at 3 h (4794.2 cm-3), which was the lowest seasonal peak (Fig. S8). Nevertheless, Nait remained at relatively high levels (4500–4800 cm-3) over an extended period from 2 to 6 h, reflecting sustained particle growth throughout the event.

Fall presented transitional characteristics, with average FR (0.28 cm-3s-1) and GR (2.35 nmh-1) higher than winter but lower than spring and summer (Table 1). Fall NPF process parameters showed transitional characteristics between summer and winter, generally similar to winter (Fig. 4c). Its FR peak (0.68 cm-3s-1) and GR peak (4.90 nmh-1) were higher than winter but much lower than spring and summer. The increase in Nait after NPF onset was 165 %, significantly stronger than in winter (Fig. S8). The Nait peak (6240.9 cm-3) occurred latest (4 h) and remained above 5600 cm-3 from 5–7 h, higher than winter.

In summary, spring shows the highest FR (7.13 cm-3s-1) but the lowest GR (3.69 nmh-1) due to a large CS, indicating strong nucleation yet suppressed growth. Summer achieves the highest GR (peak 11.68 nmh-1) and the cleanest background (lowest CS), where growth dominates over formation. Fall and winter exhibit low FR and GR with delayed growth, reflecting weaker NPF intensity.

Spring NPF days were characterized by the highest CS (4.1×10-2 s-1) and were predominantly influenced by secondary pollution. Spring NPF events were mainly driven by the continuous transport from the southeast coastal pathway (CL2, 51.33 %, Fig. S9a), which supplied high levels of NH3 (6.76 ± 2.09 µgm-3) and HNO2 (2.81 ± 2.16 µgm-3) (Table S4). Consistently, during the NPF events, the frequency of SE/S winds continuously increased and became dominant in the later stage (maximum 38.1 %), reflecting the influence of this coastal air mass at the local scale. Before NPF events, gaseous SO2 and NH3 increased from 0.47 and 5.01 to 0.89 and 5.95 µgm-3, respectively (Fig. 5a). Two hours after NPF onset, their concentrations began to decline continuously, indicating substantial consumption. Mass concentrations of secondary inorganic salts (SO42- and NO3-) fluctuated between 4.5–4.8 and 4–6 µgm-3, respectively. Although NO3- showed minor fluctuations, it remained at relatively high levels (Fig. 6a). NPF events occurred when the temperature was between 18.9–23.2 °C (Fig. S12a), favoring accelerated photochemical reactions and gas-particle conversion of semi-volatile gases (Chen et al., 2023). Wind speeds were generally low (<1.9 ms-1), and the atmospheric stratification was stable. After NPF onset, O3 increased significantly (from 81.1 to 98.8 µgm-3; Fig. S14a), indicating enhanced atmospheric oxidizability. Precursor gases (SO2 and NO2) were oxidized via photochemistry to form NO3- and SO42-, promoting the generation of secondary inorganic salts and secondary aerosols, including secondary organic aerosol (Fig. S13a). The high CS background, spring had high-frequency and high-FR NPF events, primarily attributed to higher precursor gas and strong photochemistry.

However, high CS competed for condensable vapors and scavenged newly formed particles, suppressing the growth stage of new particles. The high particle hygroscopicity (0.56) was observed at -4 h before the event, and it decreased to 0.53 during the NPF event (1–4 h) (Fig. S11).

CS in summer, fall, and winter were relatively low (around 2.0×10-2 s-1), indicating fewer surfaces available for condensation in the atmosphere.

Summer NPF events occurred under a high-temperature environment (>32 °C) (Fig. S12b), which inhibits nucleation (Yu et al., 2017). Two hours after NPF onset (2 h), the frequency of northeasterly (NE) winds surged from 0 % (at -4 h) to 50 %, corresponding to marine air mass cluster CL2 (60 %), which had a low average CS of 1.08 ± 0.31×10-2 s-1 (Table S3). The arrival of this air mass diluted local pollutants, causing PM₁₀ to drop from 14.25 µgm-3 at 0 h to 8.75 µgm-3 at 2 h; meanwhile, it supplied abundant gaseous precursors or enhanced local photochemical activity, leading to continuous increases in SO2 and NH3 from 0.48 and 4.00 µgm-3 to peak values of 0.94 and 5.18 µgm-3, respectively, at 2 h (Fig. 5b). Therefore, despite favorable conditions of low CS and ample sunlight, summer NPF frequency and formation rate were low. As ozone continued to rise (0–6 h), secondary ions (NO3-, SO42-) and SOC increased (Figs. 6b and S13b). These efficiently condensed onto particle surfaces under a low CS background, achieving extremely high growth efficiency (GR peak of 11.59 nmh-1). κinorg in the range of 0.61–0.75 was higher than in other seasons, surging to 0.75 at 3 h after NPF onset. This increase was primarily driven by prevailing easterly and southerly winds, corresponding to air mass cluster CL3, which had a high κinorg (0.71). This indicates that inorganic salts of marine origin facilitated the sustained growth of particles.

Before Fall NPF events, SO2 and NH3 increased from 0.36 and 1.99 to 2.83 and 2.83 µgm-3, respectively (Fig. 5c). This stage was dominated by northwesterly winds (NW: 62 %) and strongly influenced by continental air masses from CL2 and CL3, which contributed to the initial accumulation of precursor gases (SO2 and NH3). The lower temperature (around 22 °C) and high humidity (RH>71 %) environment (Fig. S12c) favored the combination of sulfuric acid molecules, promoting nucleation (Lehtipalo et al., 2018; Tröstl et al., 2016; Yue and Hamill, 1979). After NPF onset, although atmospheric oxidizability increased continuously (O3 from 36 to 94 µgm-3), it was short-lived. Within 1–3 h after NPF onset, the local wind direction shifted notably to southerly (S: 44.4 %, Fig. S10c), corresponding on a larger scale to the marine/coastal air mass CL1 (22.22 %, Fig. S9c). CL1 served as the most important transport pathway for NH3 in the fall, with an average concentration as high as 3.13 µgm-3 (Table S4), substantially higher than that of continental pathways. After 4 h, as O3 gradually decreased and primary emissions increased, pollutant accumulation occurred, and BC and POC rebounded (Fig. S13c). The contribution of primary emissions (e.g., BC, POC) to aerosols was significantly enhanced, and overall κinorg (0.55) was low.

Winter NPF events were also preceded by an accumulation of gaseous precursors (Fig. 5d). However, elevated emissions from sources such as heating led to high BC and POC during the initial NPF stage (1.86 and 1.02 µgm-3 at -4 h, Fig. S13d). These abundant primary particles strongly suppressed new particle formation via intense coagulation scavenging, resulting in persistently low formation rates (FR≤0.47 cm-3s-1). Photochemical activity was limited under low winter temperatures (Fig. S12d). As temperatures continued to drop later in the event (after 6 h), condensation-driven conversion of gaseous precursors to particles increased (Jokinen et al., 2018). During the 6–10 h period, the high concentrations of precursors reached supersaturation under the low-temperature conditions, promoting the rapid formation of particulate ammonium nitrate (nitrate concentration increased from 4.21 to 5.14 µgm-3, Fig. 6d). In winter, the air masses were mainly dominated by the northwesterly continental air mass originating from Siberia (CL1, Fig. S9d). Particles within this air mass exhibited relatively high κinorg (0.59, Table S3), and the introduction of these hygroscopic components effectively offset the dilution of overall hygroscopicity caused by BC and POC. During the 8–10 h period, northwesterly winds persisted, and their frequency gradually increased, which also explains the gradual increase in κinorg values (from 0.57 to 0.61, Fig. S11).

In summary, in spring, high NH3 and photochemistry drive strong nucleation, but high CS and secondary inorganic salts suppress growth. Summer marine air masses provide low CS and high hygroscopicity (κinorg>0.6), favoring efficient growth, while fall and winter continental emissions (BC, POC) lower κinorg (≤0.55) and limit NPF intensity.

According to SHAP analysis (Fig. S15), the main contributors to FR in Fuzhou are nucleation mode (76.2 %), CS (13.8 %), NH3 (1.3 %), and SO2 (0.6 %). Physical processes explain about 90 % of the variance. A prominent feature is a sharp NH3 threshold at 4 µgm-3. Below this value, SHAP is approximately -0.025; above it, SHAP becomes positive and rises to approximately 0.2 at 10 µgm-3. CS shows a clear inhibition onset at 0.03 s-1, with SHAP dropping from 0 to -0.5 as CS increases to 0.08 s-1. Temperature has a positive SHAP only within 20–25 °C (peak 0.01), decreasing by 40 % at 35 °C. As shown in Fig. 7, the strongest interaction is between nucleation mode and CS (0.40), far exceeding other pairs (NH3-nucleation mode: 0.04), highlighting that the net FR is governed by the competition between particle formation and scavenging.

For the Aitken mode, the driving forces shift notably. CS remains the top contributor, followed by accumulation mode (13.8 %) and nucleation mode (7.1 %). As shown in Fig. S16, a key distinction from FR is the temperature response. Instead of a bell shape, SHAP increases linearly from 0 to 0.1 over 20–35 °C, indicating that high temperatures accelerate particle growth, enabling a rapid transition from nucleation to the Aitken mode. The NH3 threshold at 4 µgm-3 persists, showing its continued role in particle growth. RH exerts a linear negative effect, turning SHAP negative above 60 %, suggesting hygroscopic growth or coagulation loss. In Fig. S17, the CS-nucleation mode interaction strength is 0.12, lower than in FR but still dominant, implying that strong nucleation can offset high CS losses.

In summary, the SHAP attribution reveals two distinct regimes. FR is dominated by CS, accompanied by a sharp chemical trigger (NH3>4 µgm-3) and a narrow temperature window (20–25 °C). The Aitken mode, while still influenced by CS, is primarily driven by temperature-accelerated growth (linearly increasing above 20 °C). Ammonia acts as a persistent enhancer in both stages, whereas high RH (>60 %) consistently suppresses Aitken mode concentrations.

In summer, although NPF frequency was the lowest, it had a significant enhancing effect on NCCN. E_NCCN at 0.4 % SS was as high as 1.64 (Fig. 8a). In the initial stage of NPF events (0–2 h), NCCN showed a sharp decline. At 0.4 % SS, it decreased from about 2078 to 1187 cm-3 (Fig. S18b). However, after 2 h, NCCN recovered noticeably and later returned to or even exceeded initial levels. The CCN size distribution (Fig. 8c) showed that during the event, smaller CCN (1–3 µm) decreased, while CCN of 5.5–6.5 µm increased. For instance, CCN at 6 µm increased from 2607.7 to 3147.8 cm-3. Summer NPF promoted CCN growth to larger sizes. However, the daily average NCCN on summer NPF days was lower than on non-NPF days, possibly due to the extremely low NPF frequency (4.35 %) and the sharp decline in CCN at the beginning of events, affecting the daily average.

Spring NPF events' impact on NCCN showed significant suppression. Throughout the event, NCCN at various supersaturation levels showed only weak and slow increases. NCCN (0.4 % SS, the same as below) increased from a pre-event average (-4 to -1 h) of about 2636 to 3192 cm-3 at 4 h, then decreased slowly from 4 to 6 h, with a larger decline at higher supersaturations (Fig. S18a). E_NCCN at 0.4 % SS was only 0.88. This phenomenon corresponds to the “high formation, suppressed growth” characteristic of spring NPF. Despite the explosive generation of nucleation-mode particles, severe competition under high CS severely hindered subsequent growth of new particles, preventing them from effectively growing to CCN activation sizes, resulting in a weak or even negative contribution to CCN.

In fall events, from 2 h onward, NCCN began to increase, rising from 3471 cm-3 to a maximum of 4752 cm-3, forming a high-value plateau lasting from 3 to 8 h (Fig. S18d). After the NPF event, NCCN increased significantly across the 1–5 µm size range (Fig. 8d). Fall NPF events most effectively and broadly increased the number of particles available for cloud droplet activation in the atmosphere. Although E_NCCN was lower due to the modest initial stage, the post-event enhancement effect was significant. This evolution process is consistent with the sustained particle growth process in fall NPF, allowing new particles to grow steadily to CCN activation sizes.

In winter NPF events, NCCN was low in the early stage (0–4 h), with insignificant growth. However, from 5 h onward, NCCN growth became significant, increasing from 4483 to 6173 cm-3 and remaining stable at high levels (Fig. S18d). During the event, changes in the CCN size distribution were not obvious. However, after the NPF event, CCN in the 2–5 µm size range showed the most significant growth (e.g., at 2 µm, from 7073.0 to 9045.1 cm-3; Fig. 8e).