1 Introduction
Although ultrafine particles (UFPs) are a subset of PM
2.5, conventional PM
2.5 measurement methods are often ineffective in characterizing these aerosols with diameters below 100 nm. As shown in Fig. 1(a), UFPs dominate particle number concentration (PNC) yet contribute negligibly to mass concentration (PMC). Consequently, PNC is a more suitable metric for quantifying UFPs. Moreover, their weak light-scattering intensity, depicted in Fig. 1(b), renders conventional optical techniques ineffective for UFP investigation (
Vasilatou et al., 2021). Since these aerosols exert both climate forcing and human health effects (
Chalupa et al., 2004;
Kuula et al., 2020;
Moreno-Ríos et al., 2022), it is crucial to develop and deploy advanced techniques for effective atmospheric UFP observation.
Currently, there is a growing international consensus on the need to monitor PNC, with specific observational requirements proposed to support health risk assessment. The World Health Organization (WHO) recommends continuous measurement of particle number size distribution (PNSD) above 10 nm at ambient monitoring stations (
WHO, 2021). The European Union mandates that member states establish sampling points to monitor potential high UFP levels for every five million residents, and designate a monitoring supersite for background levels of urban air pollutants, including UFPs, for every ten million residents (European Parliament & Council of the European Union, 2024). Existing aerosol monitoring networks provide a fundamental infrastructure for observing UFPs (
Birmili et al., 2009;
Kivekäs et al., 2009;
Li et al., 2024). Moreover, mobile monitoring and numerical modeling provide distributions of PNC with high spatial resolution for assessing local emissions and population exposure (
Steffens et al., 2017;
Lloyd et al., 2024).
Beyond PNC and PNSD, a comprehensive inves-tigation into properties of ambient UFPs is necessary for the accurate assessment of their exposure risks and environmental impacts. These properties include chemical composition, surface area, hygro-scopicity, morphology, and density. Measuring key chemical components of UFPs is directly relevant to both source identification and toxicity analysis (
Pietrogrande et al., 2022;
Li et al., 2023b). Particle surface area has been suggested as one of the most relevant metrics for evaluating health impacts (
Schmid and Stoeger, 2016). Measurements of particle density and morphology are critical for refining lung deposition modeling based solely on diameters (
DeCarlo et al., 2004). Measure-ments of particle hygroscopicity are essential for accurately characterizing climate-relevant effects of UFPs, such as contributing cloud conden-sation nuclei (
Dutta et al., 2023).
To date, numerous studies have reported on the prin-ciples and technological developments of aerosol measurement (
McMurry, 2000b;
Zhang and Han, 2025a,
2025b).
Morawska et al. (2009) compre-hensively reviewed early UFP measurement techniques.
Lei et al. (2024) compiled current analytical techniques for UFP and reviewed their applications in assessing UFP health and climate impacts.
Li et al. (2023a) reviewed the techniques for online analysis of UFP chemical composition and the challenges faced. This paper further reports on recent advances in measure-ment techniques. It also reviews current methods for characterizing atmospheric UFPs, identifies remaining challenges, and discusses potential solutions involving machine learning to develop a comprehensive under-standing of atmospheric UFPs.
2 Measuring properties of UFPs
2.1 Particle number concentration
PNC is a key regulatory parameter for particle emissions, and can be measured by electrical methods and optical counting (
Flagan, 1998;
Abdillah and Wang, 2023). In electrical approaches, PNC is measured by detecting the current generated by charged particles within a Faraday cup. The accuracy of this method fundamentally depends on an accurately quantified particle charging state, which is highly variable and difficult to determine, as it hinges on the particles, the ambient conditions, and the charger itself (
Wiedensohler et al., 1986;
Tigges et al., 2015). These limitations may lead to measurement uncertainty compared to other methods (
Fierz, 2011).
Optical detection method counts particles by directly capturing their light scattering signals, typically covering a measurable size range of 0.3–10 µm (
Iida and Sakurai, 2018). By coupling particle condensation growth technology, a condensation particle counter (CPC) can detect nanoparticles (
McMurry, 2000a). Adjusting the temperature difference between the saturator and condenser sections (
Baltzer et al., 2014;
Attoui, 2018), and employing working fluids with high surface tension and low vapor pressure (
Iida et al., 2009;
Jiang et al., 2011) facilitate the reduction of the minimum detectable size to ~1 nm. Recently, a convertible CPC has been developed to accommodate different working fluids, thereby improving operational flexibility (
Li et al., 2025). Optimizing the optical and circuit designs to minimize particle pulse width can mitigate the coincidence and shielding effects that cause PNC underestimation at high concentrations, thus raising the CPC's upper detection limit (
Wang et al., 2020).
2.2 Particle number size distribution
PNSD can be used to investigate atmospheric physico-chemical processes (
Deng et al., 2020), such as new particle formation (NPF) and growth due to condensation and heterogeneous reactions, as well as for source apportionment (
Beddows et al., 2009). Particle size can be characterized by different metrics, such as aerodynamic diameter and mobility diameter. These metrics not only differ in their measurement principles, but also often yield distinct values for the same particle (
McMurry, 2000b).
Aerodynamic diameter determines particle aerody-namic behavior, such as deposition in the human respiratory tract (
Bartol et al., 2024). Impactors separate and sample UFPs of different aerodynamic sizes based on inertia effects (
Sardar et al., 2005;
Zahir et al., 2025). Recently, the aerodynamic aerosol classifier (AAC) was developed. This instrument classifies particles by aerodynamic diameter based on the equilibrium between centrifugal force and drag force in a rotating flow field (
Tavakoli and Olfert, 2013). Currently, commercial AAC instruments can classify particles with aerodynamic diameters down to ~25 nm (
Pongetti et al., 2022).
Electrical mobility, defined as the terminal velocity of a particle in unit electric field, is the key metric utilized in mobility size classifiers (
Hinds and Zhu, 2022). Common instruments, such as the Differential Mobility Analyzer (DMA), classify particles by varying the electric field strength. Only charged particles within a specific mobility diameter range can reach the sampling slit of the DMA for a given voltage setting, and be counted. These counts are then corrected for particle charging fraction, CPC detection efficiency, system transmission efficiency, and DMA transfer function to derive the size-resolved PNC. The Scanning Mobility Particle Sizer (SMPS) is the commonly used instrument designed based on this principle for measuring PNSD. A typical SMPS operates in the size range of approximately 10–1000 nm with a time resolution of 1–2 minutes. Additionally, the Fast Mobility Particle Sizer (FMPS) employs an integrated multi-channel electrometer electrode array to simul-taneously measure currents from charged particles of different sizes, achieving a time resolution of 1s across a size range of 5.6–560 nm (
Tammet et al., 2002). To observe nanometer-sized UFPs or molecular clusters, specially designed instruments are required. Nano-DMA (1–150 nm) or the Neutral Aerosol Ion Spectrometer (NAIS) (0.8–40 nm) can be used for mobility analysis (
Chen et al., 1998;
Manninen et al., 2016;
Cai et al., 2017).
Assuming fixed particle charging fractions are widely used for PNSD inversion but may introduce uncertainties (
Fuchs, 1963;
Wiedensohler, 1988;
Wiedensohler et al., 2012). This is because the actual charging process is dynamic, not steady-state, leading to significant inaccuracy in charge fraction estimation (
Jiang et al., 2014). Since the relative differences in positive and negative ion characteristics are critical factors influencing the bipolar charging fractions (
Jiang et al., 2014;
Tigges et al., 2015), both positively and negatively charged particles are recommended to be accounted for during the measurement and the data inversion (
Jiang et al., 2014). Thus, one can achieve true particle charging fractions in real-time and reconstructs the original PNSD based on the combined proportions of positively and negatively charged particles (
Gunn and Woessner, 1956;
Chen and Jiang, 2018;
Chen et al., 2018). This approach not only significantly improves inversion accuracy, but also enhances the robustness of measurements against variations in aerosol charging states (
Li et al., 2022b,
2024).
2.3 Surface area and density
Surface area, a proxy for the chemical reactivity of particles, can be derived from the measured diameter by assuming sphericity (
Gong et al., 2019). Oftentimes, however, particles are not spherical (
Lv and Zhao, 2025). This morphological deviation becomes particularly critical in studies of aerosol dynamics and pathological investigations (
Voth and Soldati, 2017;
Lv et al., 2024), necessitating more accurate charac-terization methods. Different measurement principles and objectives lead to varying definitions of particle surface area. The Brunauer-Emmett-Teller (BET) surface area, determined through low-temperature nitrogen adsorption, characterizes the total physical specific surface area including pore structures (
Haul, 1982). The active surface area, quantified by the adsorption kinetics of radiolabeled tracer molecules, reflects the effective interface participating in specific chemical reactions (
Konstandopoulos et al., 2004). In health effect studies, the lung deposited surface area (LDSA) serves as an indicator of particle reactivity and lung interaction potential (
Edebeli et al., 2023). It can be directly measured by electrical current detection following corona charging under simulated respiratory conditions (
Fissan et al., 2007), or derived by applying a weighting integration to the PNSD based on established alveolar deposition fraction curves (
Bair, 1992).
Particle density, defined as true or effective, serves as the bridge between mobility and aerodynamic diameter. The true density is determined based on the fundamental mass-to-volume relationship. The volume of particles is typically measured via displacement by inert gas (
Bartley et al., 2020). This volume is combined with mass measurement to obtain the true density. This method significantly eliminates the influence of particle morphology on the volume measurement, providing the intrinsic material density. The measurement of effective density is commonly based on two methodological approaches. The first approach couples a Centrifugal Particle Mass Analyzer (CPMA) in tandem with a DMA to classify the mass of mobility-selected particles. With particle volume estimated from the mobility diameter, the effective density can be calculated (
Gu et al., 2025). The second approach infers particle density by combining measurements of aerodynamic diameter and mobility diameter (
Hering and Stolzenburg, 1995).
2.4 Morphology
The weak light scattering of UFPs falls below the detection limit of conventional optical instruments designed for morphological analysis (
Kaye et al., 1996;
Burton et al., 2012;
Ding et al., 2016). Consequently, morphological analysis of UFP samples is typically conducted using off-line electron microscopy (
Dong et al., 2019;
Kirpes et al., 2020). Based on analysis of high-resolution particle images, morphology can be characterized either through direct description or calculating specific shape factors.
While direct morphological description accurately captures features, its lack of quantifiable metrics hinders integrating them into aerosol dynamics analysis. Lv and Zhao have summarized shape factors used to quantify particle morphology (
Lv and Zhao, 2025). Among these, sphericity is often used, defined as the ratio of the surface area of a sphere with the same volume as the particle to the actual surface area of the particle (
Wadell, 1932). This metric effectively quantifies the deviation of a particle's shape from a perfect sphere and has employed in simulations of the aerodynamic behavior of non-spherical particles.
2.5 Chemical components
Particle compositional information serves as a tracer for sources and a predictor of toxicity. Well-established laboratory analytical procedures provide comprehensive chemical characterization for aerosol samples (
Morawska et al., 2009). However, these offline methods cannot resolve the rapid atmospheric aging of UFPs (
Li et al., 2022a).
Online analytical techniques vary according to the targeted level of detail in chemical characterization. For specific components analysis, typical instrument combinations are often required. For example, ultrafine black carbon (BC) analysis relies on a BC detector coupled with a DMA (
Ning et al., 2013). By inserting a thermodenuder between tandem DMAs, the volatility distribution of aerosol components can be determined from particle size shrinkage after thermal desorption (
Rader and McMurry, 1986;
Sakurai et al., 2005;
Oxford et al., 2019). Based on this information, the aging state of ambient aerosols and the mixing state evolution of newly formed particles can be further analyzed. (
Chen et al., 2022;
Wu et al., 2023). Further chemical characterization requires mass spectrometric techniques. For instance, high-resolution time-of-flight aerosol mass spectrometer (HR-ToF-AMS) can quantitatively resolve major species including sulfate, nitrate, and ammonium, as well as organics and their elemental composition (
Hu et al., 2016). For detailed time-resolved near-molecular-level characterization, the thermal desorption-chemical ionization mass spectro-meter (TDCIMS) is a commonly used technique, owing to its relatively low fragmentation and high sensitivity (
Smith et al., 2004). A dedicated review of efforts to advance real-time analysis of atmospheric UFP molecular composition is available in the literature (
Li et al., 2023a).
2.6 Hygroscopicity
Conventional SMPS measurements often lack information on hygroscopic growth due to pre-drying, leading to discrepancies between measurement and reality (
Fajardo et al., 2016;
WMO, 2016). Hygro-scopicity measurements are thereby critical for assessing hygroscopic growth, which directly influ-ences particle’s climate and health effects (
Vu et al., 2015;
Jung et al., 2023a). By classifying monodisperse particles using a DMA and monitoring their size or state after hygroscopic growth, the hygroscopic growth factor (GF) and hygroscopicity parameter (
κ) can be determined (
Rader and McMurry, 1986).
An alternative approach derives particle’s hygro-scopicity from its chemical composition. By applying the Zdanovskii-Stokes-Robinson (ZSR) rule, the overall hygroscopicity parameter
κ can be estimated as the mass-weighted average of the
κ of individual components (
Zdanovskii, 1936;
Stokes and Robinson, 1966). For instance, Zhao et al. determined the size-resolved chemical composition of particles (10 nm–18 μm) and well derived
κ values based on ZSR rule (
Zhao et al., 2020). Conversely, based on the hygro-scopicity parameter
κ and the ZSR rule, the volume fractions of major chemical species can be estimated as well (
Aklilu and Mozurkewich, 2004). When combined with volatility analysis, the internal mixing state of particles can be further analyzed (
Villani et al., 2008).
2.7 Toxicity
Oxidative stress, resulting from reactive oxygen species (ROS) generated by redox-active components, is a pivotal toxicological mechanism for aerosol health effects (
Jiang et al., 2008;
Gao et al., 2020). The ROS-generating capacity is defined as oxidative potential (OP), quantified via acellular assays such as the dithiothreitol (DTT) method, which measures OP by the DTT consumption rate in simulated lung fluid (
Kurihara et al., 2022). Given the abundance of redox-active constituents in UFPs, OP is likely key to understanding UFPs health effects (
Chen et al., 2016).
While mass-normalized OP assays provide valuable insights into the intrinsic toxicity of UFP compared to PM
2.5 (
Jeng, 2010), they cannot fully address the paradox of how such low mass concentrations exert significant adverse health effects. Resolving this question requires a more systematic, biologically grounded approach.
In vivo studies using animal models expose subjects to UFPs via direct inhalation or systemic injection, inducing biological outcomes and elucidating pathogenic pathways (
Farina et al., 2019;
Saleh et al., 2019), followed by integrated analysis of physiological parameters and target organ histo-pathology. However, the translational applicability of these studies faces several inherent limitations, including variability in exposure conditions and fundamental physiological disparities between animal models and humans (
Jiang et al., 2009).
3 Characterizing atmospheric UFPs
UFPs exhibit high spatial and temporal variability, making long-term multi-site observations essential for characterizing their regional distribution. However, current long-term UFP monitoring networks primarily focus on PNC and PNSD. Chemical compositional analysis largely relies on offline analysis or short-term campaigns (
Nordmann, 2009;
Crenn et al., 2015). For other properties, such as hygroscopicity, our under-standing relies on historical data compiled from multiple independent studies (
Zou et al., 2025). Given that methods for comprehensively characterizing UFP properties remain under development, this review focuses specifically on techniques for characterizing PNC and PNSD.
As shown in Fig. 2, while network monitoring provides long-term PNC data over large regions, it lacks the spatial detail to resolve local variations. Conversely, mobile platforms capture PNC at high spatial-resolution, but are confined to limited areas. Combining with other simultaneous observation, such as ancillary pollutants and meteorology, these results can be further used to establish predictive models, enabling high spatiotemporal resolution (e.g., ~1 km and ~1h) mapping of UFP.
3.1 From fixed measurements to network monitoring
As summarized in Table 1, the particle size ranges measured vary substantially across different UFP monitoring networks, with inconsistencies also observed between sites within the same network. Although a lower size limit of 10 nm is recommended (
WHO, 2021), there is no broad scientific consensus on this value, which partially explains the observed inconsistencies. Nevertheless, this significantly compromises the data comparability across various sites (e.g., PNC varies significantly with the lower size limit). While lognormal fitting can estimate PNSD below the instrumental lower detection limit (
Whitby, 1978), such extrapolation may substantially under-estimate particle number concentrations during NPF events. For specific research needs, the Global Atmosphere Watch (GAW) program recommends a detection limit of 3 nm (
WMO, 2016). To ensure data comparability, a 10 nm lower size limit should be applied as a minimum requirement. This aligns with conventional SMPS capabilities.
3.2 Mobile observation of UFPs
By assigning data to short segments (typically 100 m in length), mobile observation can generate high-spatial-resolution PNC maps along driving routes (
Kerckhoffs et al., 2021;
Doubleday et al., 2023). This approach is particularly valuable in complex urban areas, allowing detailed assessment of UFP impacts on ambient air quality and population exposure (
Weichenthal et al., 2014).
Mobile measurements require high time resolution, thereby limiting instrument selection. Since a vehicle passes a 100-m segment in only a few seconds, the instrument must complete multiple measurements within this brief time window to ensure data robustness. Consequently, conventional SMPS, with a minute-level time resolution, is unsuitable for mobile PNSD measurement. Alternatively, FMPS have been employed to obtain high time resolution PNSD data under these conditions (
Weimer et al., 2009). Besides, many studies focus solely on measuring PNC during mobile surveys (
Gani et al., 2021;
Kerckhoffs et al., 2021). Portable instruments, such as the Discmini or P-Trak, are another common choice for particle counting, as their compact and low-power consumption is well-suited to the constraints of mobile platforms (
Doubleday et al., 2023;
Lloyd et al., 2023). In contrast, the SMPS is feasible for PNSD measurements in aerial observations (
Wehner et al., 2010;
Liu et al., 2024). The platform’s ability to hover for extended periods accommodates the longer scan time of the SMPS.
3.3 From in-situ measurements to regional UFP mapping
Statistical relationships between PNC and influencing factors can be utilized to build predictive models for regional UFP mapping (
Crippa et al., 2013;
Reggente et al., 2014). Among these, Land Use Regression (LUR) modeling is widely used, as comprehensively reviewed by Hoek et al (
Hoek et al., 2008). However, developing robust models requires substantial input data, which is often obtained through resource-intensive mobile monitoring campaigns or short-term measure-ments at a large number of sites (
Kerckhoffs et al., 2016,
2017;
Jones et al., 2020).
The advancement of machine learning (ML) technology has further propelled the development of predictive models. As shown in Table 2, ML models generally perform well in predicting PNC across various atmospheres, as indicated by high coefficients of determination (
R2). Vachon et al. reported that ML models, especially tree-based models, generally outperform conventional statistical models on the same dataset (
Vachon et al., 2024a,
2024b).
A striking inconsistency exists in the literature regarding the selection of key predictors for ML-based UFP models. Jung et al. identified “surface pressure” and “distance to road” as the most important predictors for modelling UFP distribution in central Taiwan, China (
Jung et al., 2023b). Xu et al reported NO
2 as the most important predictor for modeling in Toronto (
Xu et al., 2022). While such discrepancies could be partly attributed to differences in UFP observation methods, they strongly indicate that the relevance and predictive power of specific predictors are highly location-dependent. Therefore, systematic investigation into governing parameters across diverse environments is needed to develop models that are both robust and parsimonious.
Existing ML prediction models primarily use PNC as the outcome variable, often overlooking PNSD. While these models demonstrate reasonable performance over long temporal scales, their predictive accuracy deteriorates significantly for short-term PNC variations, such as predicting daily-average value (
Rahman et al., 2020;
Jianyao et al., 2025). The underlying drivers for increased PNC can vary substantially across different size ranges (
Beddows et al., 2009). A generalized parameterization of the relationship between total PNC and predictors may fail to capture these distinct dynamics, leading to inaccuracies in short-term forecasts. Rahman et al. reported considerable variation in the importance of predictors when modeling PNC for different size ranges (
Rahman et al., 2020), thereby recommending the disaggregation of total PNC into its modal components for model training and prediction purposes.
4 Future outlook
While extensive worldwide monitoring campaigns for PNC and PNSD have provided a good data foundation for training ML models, the poor comparability of data across different campaigns poses a latent risk of introducing hidden biases that undermine model robustness over large scales. In addition to the varying measured size ranges mentioned in Section 3.1, PNC readings from the Discmini, commonly used in mobile monitoring, can be 10%–20% lower than SMPS (
Fierz et al., 2011;
Mills et al., 2013). This deviation can remain hidden during model evaluation, as the “ground truth” is unavailable for validation. To improve dataset accuracy beyond periodical instrument calibration and inter-instrument comparison,
Wiedensohler et al. (2012,
2018) recommended establishing technical standards for SMPS measurements, along with data protocols that ensure traceability throughout the data inversion. These harmonization efforts, however, require updates to accommodate new progresses that replace unrealistic fixed charge fractions with dynamic charge fractions. It has been demonstrated that dynamic charge fractions can be continuously revealed and used for SMPS data inversion (
Li et al., 2022b,
2024).
The WHO recommends defining low PNC as <1000 particles/cm
3 (24-h mean) and high PNC as >10000 particles/cm
3 (24-h mean) to guide decisions on the priorities of UFP source emission control (
WHO, 2021). However, even in clean region, elevated PNC levels can also arise from NPF (
Venzac et al., 2008;
Cai et al., 2018). To inform evidence-based regulation, it is essential to comprehensively characterize UFPs across different environments including PNSD and chemical composition (
Ye et al., 2020). Although current research on UFP properties beyond PNSD remains spatially and temporally limited (
Air Quality Expert Group, 2018), the accumulation of long-term data and advances in ML offer promising pathways to overcome these barriers. Using meteorological parameters and ancillary pollutants as predictors, ML has been successfully applied to predict long-term variations in PM
2.5 chemical components (
Lv et al., 2021). Given the strong link between UFP size and its sources, incorporating PNSD as an additional predictor into such models could further enhance their robustness. This would enable the extraction of long-term spatial patterns for properties, such as chemical composition, from existing PNSD results. Such integration across different campaigns underscores the need for standardized measurement protocols for UFP properties beyond PNSD, as well as for systematic inter-instrument comparisons, which has been highlighted by well-recognized instrumental discrepancies in mass spectrometry (
Riva et al., 2019). Ultimately, compre-hensive UFP characterization, coupled with in-depth analyses, such as lung deposition modeling and toxicological studies, can elucidate the exposure risks associated with specific size distributions and chemical components, thereby supporting the development of targeted air quality and emission control policies.
PDF
(2966KB)
