A continuous online observation campaign was conducted at the Shanghuang Ecological and Environmental Observation of the Chinese Academy of Sciences (Shanghuang station; 28.58° N, 119.51° E) from 19 April–30 May 2024 (Fig. S1 in the Supplement). The station is located in Jinhua City, Zhejiang Province, at an elevation of 1128 m above sea level (a.s.l.). It is characterized by mountainous terrain and forest coverage, representing a typical high-altitude background environment in the YRD region of China, more details about Shanghuang station could be found in Zhang et al. (2024) and Wang et al. (2025).

Ambient particles and droplets were initially selected using an advanced aerosol–cloud sampling inlet system, which alternated between the PM1 cyclone, PM2.5 cyclone and total suspended particulate (TSP) passage every 20 min (Xu et al., 2024). To minimize particle loss within the sampling system, the relative humidity (RH) at the inlet was maintained below 30 % using a Nafion dryer and a sheath air cycle system. Additionally, diffusion and gravitational losses in the inlet tubing were corrected based on the tubing shape and flow rate (Baron and Willeke, 2001). The particle number size distribution (PNSD) from 2.5 nm–20 µm was continuously measured using an integrated system. The system consisted of a Neutral Cluster and Air Ion Spectrometer (NAIS, Airel Ltd.) covering a mobility diameter (dm) range of 2.5–42 nm, a scanning mobility particle sizer (SMPS, model 3936, TSI Inc.) for 14.5–710 nm (dm) comprising a model TSI3080 electrostatic classifier and a model TSI 3775 condensation particle counter, and an Aerodynamic Particle Sizer (APS, model 3221, TSI Inc.) for 0.5–20 µm (aerodynamic diameter, da). Prior to and during the campaign, regular zero checks and flow-rate verifications were performed using a calibrated primary flow meter. The NAIS was operated at a sample flow rate of 60 Lmin-1 to minimize diffusion losses, with data recorded at 10 min resolution (Mirme and Mirme, 2013). The SMPS was run with an aerosol-to-sheath flow ratio of 0.3 : 3.0 Lmin-1 (1:10), and the APS with an aerosol flow of 1.0 Lmin-1 and a sheath flow of 4.0 Lmin-1 (Liu et al., 2016). Data from the SMPS and APS, recorded at 5 min resolution, were averaged into hourly spectra and merged into a unified particle size spectrum matrix (dm: 14.5–16 000 nm) following the procedure described by Beddows et al. (2010).

To explore the chemical difference of newly formed particles during their growth processes, the volatile characteristic of those particles was measured using the Thermal Denuder (TD)-SMPS system. Volatile analysis helps distinguish between categories of inorganic compounds, such as nitrates and sulfates and can indirectly provide information on aerosol composition (Schmid et al., 2002). By comparing the volumes of heated (300 °C) and unheated particles, the volatility characteristics of particles are studied under the assumption that the particles are spherical and characterized using the volume fraction remaining (VFR) of submicron aerosols. The remaining semi-volatile components are mainly organic compounds, while components such as sulfates and nitrates are evaporated at high temperatures, thereby investigating changes in the proportion of semi-volatile components during the NPF growth process.

CCN number concentration can be measured by CCN counter (Model CCN-100; Deng et al., 2011). The instrument operated at five supersaturation – 0.07 %, 0.11 %, 0.20 %, 0.40 %, and 0.80 % – with each level maintained for 15 min. To ensure data reliability, measurements were filtered based on established quality control criteria addressing SS instability within the growth chamber (Rejano et al., 2021). The total flow rate was maintained at 0.5 Lmin-1, with an aerosol to sheath flow ratio of 1:10.

The chemical composition of non-refractory submicron particles (NR-PM2.5), including organics, sulfate, nitrate, ammonium, and chloride, was measured using an Aerodyne Time-of-Flight Aerosol Chemical Speciation Monitor (ToF-ACSM, Li et al., 2023). The instrument sampled ambient air through the same inlet as the PNSD system, with a flow rate of 0.1 Lmin-1 and a time resolution of 10 min. The ToF-ACSM was operated with a capture vaporizer, and its ionization efficiency (IE) was calibrated at the start of the campaign using 300 nm ammonium nitrate particles. The default relative ionization efficiencies (RIEs) for nitrate, organics, and chloride (1.1, 1.4, and 1.3, respectively) were applied (Nault et al., 2023). According to the ion efficiency (IE) calibration results using ammonium sulfate, the RIE values of ammonium and sulfate were 5.05 and 0.73, respectively (Zhang et al., 2024). A composition-dependent collection efficiency (CE) was applied to the raw data to correct for particle losses in the aerodynamic lens, following the parameterization established by Middlebrook et al. (2012).

The concentrations of major gaseous precursors were measured using the following commercial analyzers: a pulsed UV fluorescence analyzer (Thermo Scientific, Model 43i) for sulfur dioxide (SO2), a UV photometric analyzer (Thermo Scientific, Model 49i) for ozone (O3), a chemiluminescence analyzer (Thermo Scientific, Model 42i) for nitrogen oxides (NOx), and a cavity ring-down spectrometer (Picarro, Model G1103) for ammonia (NH3). Prior to the campaign, all gaseous analyzers (SO2, O3, NOx, and NH3) were calibrated with certified reference gases and zero air. In addition, routine calibration checks for these gaseous instruments were performed biweekly throughout the measurement period to ensure continuous accuracy and consistency of the data of gaseous pollutants. Additionally, PM2.5 mass concentrations were measured using a continuous ambient particulate monitor (Model 5014i, Thermo Scientific, USA), with a PM2.5 size cut-off applied prior to the sampling inlet. Meteorological parameters were recorded during the measurement period using an automated weather observation system (Milos520, Vaisala, Finland) positioned adjacent to the PNSD system. A more comprehensive description of the instruments is available in our previous work (Yang et al., 2021).

Based on their size, atmospheric aerosol particles are commonly grouped into four modes: nucleation mode (< 20 nm), Aitken mode (20–100 nm), accumulation mode (100–1000 nm), and coarse mode (> 1 µm). In this study, the number concentration of each mode was obtained by integrating the measured particle number size distribution over the corresponding diameter interval. An NPF event is identified when a distinct and sustained (≥ 2 h) burst of nucleation-mode particles – particularly in the sub-6 nm size range – is observed, followed by a clear growth of the mode to larger sizes (Dal Maso et al., 2005). Cases failing to meet these criteria were classified as non-NPF events. The formation rate (J2.5), growth rate (GR) and CS were calculated with the commonly used method (Yang et al., 2021). Recognized as a key contributor to particle nucleation, the concentration of sulfuric acid (H2SO4) was estimated via a proxy approach proposed by Lu et al. (2019). Additionally, to assess how sulfuric acid (H2SO4) influences the early-stage particle growth, its contribution to the initial growth rate was quantitatively evaluated using the equation introduced by Nieminen et al. (2010).

In this study, the NCCN and activation diameter (Da) was calculated by κ-Köhler theory (Petters and Kreidenweis, 2007), which simply link the Da with the supersaturation, is applied as follows, when κ>0.1: 1κ=4A327Da3ln⁡2S2A=4σωMωRTρω Here, σω denotes the surface tension of the droplet at the activation point (σω=0.072Jm-2), Mω is the molecular weight of water (Mω=0.018015kgmol-1), T is the temperature of the air parcel, R represents the universal gas constant (R=8.315JK-1mol-1), and ρω refers to the density of water (ρω=997.1kgm-3). The hygroscopicity parameter κ, which reflects the water affinity of aerosols, is influenced by their chemical composition. In this study, κ was estimated using the Zdanovskii–Stokes–Robinson (ZSR) mixing rule (Stokes and Robinson, 1966), based on chemical volume fractions under the assumption of internally mixed particles, following the approach of Gunthe et al. (2011), as follows: 3κchem=∑iεiκi Here, κi and εi represent the hygroscopicity parameter and volume fraction of each individual dry component in the mixture, respectively. The κ values and corresponding densities (ρ) used in the calculations were adopted from Petters and Kreidenweis (2007) and Topping et al. (2005). While the approach combining the critical dry diameter and bulk aerosol properties may introduce some degree of uncertainty, previous studies have shown that the discrepancy between predicted and measured CCN concentrations remains within an acceptable margin of ±20% under both polluted and pristine atmospheric conditions (Zhang et al., 2017).

During the intense campaign, eight NPF events were identified across 39 valid observation days from 19 April–30 May at the Shanghuang station. Note that bursts in the concentration of freshly nucleated sub-6 nm particles were also observed on 30 April and 16 May (see Fig. 1d). However, these two episodes were not classified as NPF events because they occurred at night and were not followed by sustained growth of the nucleation mode to larger sizes, which is a key criterion for defining a full NPF event. Meteorological elements (Fig. 1e and f) show that southerly winds dominated during the period of observation, with low average wind speeds (1.9 ms-1). The average relative humidity (RH) during the NPF occurrence time (07:00–18:00 LT) was 63 % and 75 % for NPF days and non-NPF days, respectively, while the temperature was comparable for NPF and non-NPF days (21.0 and 19.7 °C). Thus, there was one NPF event in April and seven in May, resulting in an overall NPF frequency of 21 %. This value is higher than the observational values for European high-altitude sites (900–1200 m a.s.l.) during the springtime (Zugspitze Schneefernerhaus: 3 %; Hohenpeißenberg: 7 %; Sun et al., 2024), while it is similar to nearby urban site (Shanghai: 20 %; Xiao et al., 2015) and mountain site in North China (Mountain Tai: 21 %; Lv et al., 2018).

The air mass clustering analysis via backward trajectories (Draxler and Hess, 1998) was performed to track the origination of these NPF events, which identified four distinct air mass categories during the observation period (Fig. S2). As showed in Fig. S1, Cluster 1 represents the polluted air masses affected by North China Plain urban emissions, Clusters 2 and 4 represent the relative clean air mass affected by western and southern urban emissions, Cluster 3 represents the air masses affected by coastal emissions. Combining trajectory analysis with PM2.5 mass concentrations, we categorized the eight NPF events into two types: NPF-C events (occurred under clean conditions in Cluster 2–4) and NPF-P events (occurred under polluted conditions in Cluster 1), with average PM2.5 of NPF-P events 101 % higher that during NPF-C events (12.8 µgm-3 vs. 6.4 µgm-3). As showed in Fig. 1d, significant variations in 2–6 nm Nucleation mode particles were observed among the eight NPF events, the peak value of which ranged from 246–1318 cm-3. The average PNSD during NPF-C events and NPF-P events were fitted as the sum of three mode lognormal distributions (Fig. 1a and b, Hussein et al., 2005), and revealed that NPF-P events exhibited higher Aitken mode particle concentrations (3978 cm-3) than NPF-C events (1980 cm-3), while the freshly nucleated sub-6 nm particles were lower in NPF-P (575 cm-3 vs. 881 cm-3). In addition, the accumulation mode particles were much higher in NPF-P than in NPF-C (881 cm-3 vs. 575 cm-3). These results indicate that NPF-P event is primarily influenced by regional transportation, whereas NPF-C reflects the background atmospheric conditions at the BLT of the mountain site.

It is worth noting that the NPF event observed on 5 May (NPF-C) occurred during a cloud interstitial period under persistently high relative humidity (> 90 %), accompany with a slightly higher formation rate (J2.5=0.8cm-3s-1) and growth rate (GR=5.7nmh-1) compared with the average value of the other NPF-C events (Table 1). We hypothesize that aqueous-phase chemical processes within the preceding cloud were pivotal. A mechanism analogous to the “post-fog growth” reported in the Arctic may be at play, whereby in-cloud reactions generate semi-volatile organic compounds (SVOCs) that later condense onto particles (Kecorius et al., 2023). While direct measurements of the specific SVOCs are not available, the elevated concentration of isoprene – a key biogenic precursor – on that day (0.3 ppbv compared to the 0.2 ppbv average for other NPF-C events) provides indirect support for enhanced biogenic activity and potential secondary organic aerosol formation pathways. Following cloud dissipation, these cloud-generated condensable vapors were released and, under sustained high humidity, rapidly condensed onto the newly formed nucleation-mode particles. This organic-dominated condensation likely surpassed the nitrate-driven growth observed in other events, facilitating sustained particle growth and enabling a larger fraction of the population to surpass the activation diameter and reach CCN sizes.

To elucidate the factors driving distinct NPF behaviors, this section presents a diurnal comparison of key parameters between clean (NPF-C) and polluted (NPF-P) event days. As shown in Fig. 2a, the average formation rate (J2.5) during NPF-P events was 2.4 cm-3s-1, approximately 3.6 times higher than during NPF-C events (0.7 cm-3s-1). The peak J2.5 in NPF-P events (6.2 cm-3s-1 at 12:00 LT) was also higher and occurred one hour later than the peak in NPF-C events (1.8 cm-3s-1 at 11:00 LT). The most pronounced enhancement – a fivefold increase – was observed at 10:00 LT (2.5 cm-3s-1 vs. 0.5 cm-3s-1). While the average gaseous sulfuric acid (H2SO4) concentration was 23 % higher in NPF-P events (8.1×106cm-3) and the condensation sink (CS) was also elevated (0.013 s-1 vs. 0.008 s-1 for NPF-C), the significantly stronger formation and growth rates indicate that enhanced production of condensable vapors from anthropogenic pollution was sufficient to overcome the increased sink strength, enabling intense NPF – a phenomenon documented in other polluted environments (Yang et al., 2021). Crucially, the 23 % difference in [H2SO4] alone cannot account for the ∼3.6-fold difference in J2.5. Ammonia (NH3) played a critical role in this enhanced nucleation. The average NH3 concentration during NPF-P events (8.1 ppbv) was approximately twice that during NPF-C events (4.1 ppbv; Fig. 2c). This elevated NH3 level, coinciding with higher H2SO4, likely contributed to the enhanced nucleation rates observed under polluted conditions by stabilizing sulfuric acid clusters.

Concurrently, NPF-P events exhibited a higher event-average of background ozone (O3) concentration (27.7 ppbv vs. 19.9 ppbv for NPF-C). Although the O3 difference narrowed during the peak nucleation period (10:00–12:00 LT) – suggesting its primary role is in maintaining an enhanced oxidative environment conducive to precursor oxidation rather than directly driving the instantaneous nucleation burst – the difference expanded again after 15:00 LT, reaching a maximum in the late afternoon (18:00 LT; Fig. 2d). This later period coincides with the sustained particle growth phase, where a stronger oxidative capacity likely facilitates the production of low-volatility condensable vapors, thereby influencing condensational growth. Correspondingly, the average particle growth rate (GR) during NPF-P events was 6.8 nm h-1, which is 24 % higher than during NPF-C events (5.5 nm h-1; Fig. 2g). The overall elevated GR is consistent with a greater abundance of condensable vapors (e.g., nitrate and photochemically generated organics), which are discussed in the following sections. Compared to typical values reported for a remote boreal forest site (Hyytiälä, Finland: J3=0.4cm-3s-1, GR=2.3nmh-1; Kerminen et al., 2018), the formation and growth rates observed at our site are higher by 275 % and 126 %, respectively. Our values are close to those reported for other Chinese high-altitude background sites like Mount Tai (J3=1-2cm-3s-1; Shen et al., 2019), Mount Heng (J15=0.15-0.45cm-3s-1; Nie et al., 2014), and Mount Yulong (J3=1.33cm-3s-1; Shang et al., 2018). These differences suggest that the intensity of an NPF event can vary significantly depending on the atmospheric conditions and the regional transport processes involved.

To investigate the chemical differences driving nanoparticle growth during the two types of NPF events, the diurnal variations of chemical components (organics, sulfates, nitrates, ammonium, chlorides, and black carbon) were analyzed during NPF evolution (Fig. 2h and i). The results show that during NPF-P events, mass concentrations of all major chemical components increased alongside particle growth, with organics and nitrates exhibiting the most pronounced and sustained enhancement (Fig. 2h and i). In contrast, NPF-C events displayed weaker and less persistent increases. While organics dominated the non-refractory PM2.5 (NR-PM2.5) mass fraction (accounting for more than half) during the growth phase in both event types, the chemical evolution pathways diverged significantly under anthropogenic influence. The stronger nitrate growth in NPF-P events can be attributed to a more favorable chemical environment. These events were characterized by significantly higher concentrations of NO2 and NH3 (Fig. 2c). Photochemical modeling indicates that elevated NO2 under stronger solar radiation leads to enhanced production of gaseous nitric acid (HNO3) (Fig. S3). In the presence of abundant NH3, this HNO3 efficiently partitions to the particle phase via neutralization, forming ammonium nitrate (Wang et al., 2022). This process explains the more than fivefold increase in nitrate peak concentrations during the later growth stages of NPF-P events, where nitrate became a key driver for sustained condensational growth. Similarly, the more substantial organic mass increase during NPF-P events is linked to enhanced secondary organic aerosol (SOA) formation (Shi et al., 2024). Higher daytime O3 concentrations (Fig. 2d) suggest a more intense oxidative environment, which promotes the photochemical oxidation of volatile organic compounds (VOCs). Coupled with elevated ambient VOC levels (e.g., isoprene), this leads to the production of more low-volatility oxygenated organic molecules that readily condense onto growing particles (Kulmala et al., 2012). Therefore, the synergistic enhancement of nitrate and organic precursors under polluted, transport-influenced conditions provides a robust chemical explanation for the faster and more sustained particle growth observed during NPF-P events compared to NPF-C events.

Previous field studies have highlighted the importance of organics for new particle growth in remote regions (Pierce et al., 2012). Recent comprehensive analyses from multiple European cities further support this view, demonstrating that the growth of nucleated particles is often driven by the condensation of semi-volatile organic compounds (Trechera et al., 2023). Our findings indicate that in anthropogenically influenced mountain regions, nitrate – primarily as ammonium nitrate (NH4NO3) – can serve as a competitive source of low-volatility condensable vapor, partially substituting for organics in driving the mass growth of new particles. This occurs under conditions of elevated NO2 and NH3, where efficient photochemical production and gas-to-particle partitioning of NH4NO3 are favored. While the strong hygroscopicity of nitrate plays a secondary role by increasing the particle's wet size (and thus potentially enhancing condensation efficiency under high relative humidity), its primary contribution to growth is through direct vapor condensation.

It should be noted that the analysis of chemical drivers for particle growth in this study relies on the bulk non-refractory PM2.5 (NR-PM2.5) composition measured by the ToF-ACSM. While CCN activation at the studied supersaturations primarily involves particles in the Aitken and smaller accumulation modes (<200nm), we assert that the bulk PM2.5 composition serves as a valid proxy for the condensing vapors during sustained NPF events under our background conditions. This is supported by the fact that during such events, the growth of the nucleation mode is the dominant source of new aerosol mass in the submicron range. Previous study indicates that changes in bulk organic and inorganic mass concentrations correlate well with the condensational needs of growing nanoparticles, making bulk composition a practical and informative metric for identifying dominant growth pathways (Vakkari et al., 2015). We acknowledge that size-dependent compositional differences may exist and represent an important avenue for future research with size-resolved instrumentation.

Gaseous sulfuric acid is recognized as an important specie in nucleation across NPF events (Garcia-Marlès et al., 2024). The correlation coefficients (R) between J2.5 and [H2SO4] were 0.77 for NPF-C events and 0.87 for NPF-P events (Fig. 3b). This positive dependence of the nucleation rate on sulfuric acid concentration is consistent with observations from remote background sites, though the strength of the correlation varies with the degree of anthropogenic influence (Kulmala et al., 2013). At pristine sites such as Hyytiälä, the correlation is often moderated by the co-involvement of biogenic organic vapors and ions (Kulmala et al., 2025), whereas at background sites in China affected by regional pollution transport, stronger correlations between nucleation and [H2SO4] was typically observed (Gao et al., 2025). The high correlations observed here (R=0.77-0.87) align with the latter pattern, reinforcing that our mountain-top station, although a background site, experiences substantial anthropogenic influence that shapes the nucleation mechanism. However, the moderate difference in [H2SO4] alone cannot explain the large difference in J2.5 between event types (Sect. 3.2). Previous studies have also indicated that binary H2SO4–H2O nucleation cannot fully account for atmospheric NPF rates (Kirkby et al., 2011). This points to the importance of additional compounds that stabilize H2SO4 clusters and modulate nucleation efficiency. Previous field and chamber studies also proposed that the gaseous species such as ammonia (Kulmala et al., 2013; Kürten, 2019) and amines (Metzger et al., 2010; Yao et al., 2018) also promote the nucleation. The elevated NH3 concentrations measured during NPF-P events (Fig. 2b) thus provide a plausible explanation for their higher nucleation rates despite a less-than-proportional increase in [H2SO4].

To explore the nucleation mechanism in the atmospheric boundary layer top, the relationship between J2.5 and [H2SO4] was analyzed for NPF-P and NPF-C events and compared with results from CLOUD chamber experiments, which delineate pathways for H2SO4–NH3–H2O and H2SO4–dimethylamine (DMA)–H2O nucleation (Kürten et al., 2019; Almeida et al., 2013). As shown in Fig. 3a, our measured formation rates (solid circles: NPF-P; hollow circles: NPF-C) fall within the [H2SO4] range spanned by these two mechanisms in the chamber. Achieving the observed J2.5 would require either higher DMA levels or higher NH3 concentrations than those set in the specific CLOUD runs. Given the lack of significant DMA sources in the region (e.g., textile or industrial activities; Chang et al., 2022), ambient NH3 (average ∼5ppbv during NPF) is the more plausible stabilizing base. However, the CLOUD experiments have not yet performed under similar atmospheric conditions as our field observation (e.g. higher NH3 levels exceed 1 ppbv) (Kürten et al., 2019). Thus, to evaluate the formation mechanism under rich-NH3 conditions representative of our site, we performed simulations using the MALTE-BOX model (Boy et al., 2006; McGrath et al., 2012), which couples the Atmospheric Cluster Dynamics Code (ACDC). Input parameters were set to the average conditions during NPF events: condensation sink (CS)=0.010s-1, [NH3]=5ppbv, RH=66%, T=293K, and pressure=883hPa. The model calculates the formation rate for clusters growing past a critical size as a function of [H2SO4]. The simulation results are shown as the yellow line and gray uncertainty band in Fig. 3a. Most of our measured J2.5 data points fall within or near the model-predicted band, indicating that H2SO4–NH3 nucleation is a quantitatively plausible mechanism under the observed conditions. The model predictions tend to be slightly higher than the measured rates. This discrepancy may arise because the model's initial cluster definition (e.g., a (H2SO4)5(NH3)5 cluster corresponding to ∼1.07nm; Huang et al., 2016) effectively simulates formation at a smaller size than our observational threshold (J2.5), and potential uncertainties in cluster binding energies or the omission of other stabilizing species (e.g., organic vapors) in the simulation. Nevertheless, the general agreement supports the conclusion that ammonia-enhanced sulfuric acid nucleation is a dominant pathway at this site.

Independent support for the role of ammonia comes from the field-observed correlations. A pronounced linear relationship exists between J2.5 and the product of H2SO4 and NH3 concentrations (Fig. 3c). The Pearson correlation coefficient (R) for J2.5 versus [H2SO4]×[NH3] ranges from 0.79–0.92, notably higher than the correlation of J2.5 with [H2SO4] alone (R=0.77-0.87, Fig. 3b). This enhanced correlation when NH3 is included as a co-variable has been observed in other polluted environments; for example, wintertime measurements in Shanghai reported a tighter relationship between J1.34 and [NH3] (R2=0.62) than with [H2SO4] (R2=0.38) (Xiao et al., 2015). Together, the consistency between our observations and the MALTE-BOX simulations, combined with the strong field-based correlation that explicitly includes NH3, provides robust evidence that ammonia plays a key role in enhancing sulfuric acid-driven nucleation at this mountain-top site.

This section aims to elucidate the relationship between the growth processes of the two types of NPF events and their efficiency in forming CCN. To quantify the CCN production from NPF events, the NCCN was calculated. Since supersaturation (SS) cannot be measured directly at the site, we employed a sensitivity approach using two representative SS values. These values were selected based on prior aircraft measurements in the regional background atmosphere, which reported a range of 0.1 %–0.5 % (Gong et al., 2023). We performed calculations for SS=0.2% and SS=0.4%, encompassing a common in-cloud condition and a higher activation threshold. For each SS, the critical activation diameter (Da) was derived using κ-Köhler theory, with the hygroscopicity parameter (κ) estimated from the measured particle chemical composition (Bougiatioti et al., 2011), adjusting for local altitude. The calculated NCCN for both SS levels was then compared with observed NCCN to evaluate the parameterization's performance and to analyze the SS-dependence of CCN production efficiency.