Abstract: The city of São Paulo, located in the southeast of Brazil, covers an area of 1,521.11 km2, with an estimated population of 11.4 million inhabitants and a vehicle fleet of approximately 8 million vehicles. This fleet is the main contributor to atmospheric pollutant emissions in the city, followed by sources such as biomass burning and the transport of pollutants from nearby industrial centers. In the context of emission control programs for conventional atmospheric pollutants implemented in São Paulo, satisfactory progress has been observed over the years. However, certain pollutant classes, such as ultrafine particles, still lack similar policies, largely due to the relatively limited research conducted on the subject. Characterizing the size and number distributions of atmospheric particles (PNSD), combined with the analysis of chemical composition and trajectories, is crucial for understanding the diverse influences of aerosols on the climate. This ongoing master’s project aims to characterize the size distributions of submicrometer aerosols in the Metropolitan Region of São Paulo. For this purpose, PNSD data were obtained using a Scanning Mobility Particle Sizer (SMPS) in the size range of 10.6 to 429.4 nanometers from June 2019 to January 2020 at the University of São Paulo. Meteorological variables and the concentrations of other atmospheric constituents, such as black carbon, ozone, and nitrogen oxides, were also monitored. Data analysis methods included algorithms for identifying modes and clustering PNSDs into different clusters. From a total of 35,703 PNSDs, log-normal curves were successfully fitted to nucleation, Aitken, and accumulation modes in 93.42%, 92.97%, and 93.39% of cases, with mean particle number concentrations (PNC) observed as (4.2 ± 2.8) x 103 (#/cm3), (5.2 ± 3.0) x 103 (#/cm3), and (2.9 ± 2.3) x 103 (#/cm3) for each mode, respectively. The average size and number distributions of particles (PNSD) for weekdays (DS) and weekends (FD) revealed a slightly smaller geometric mean diameter in DS, with an average of 49±2 nm, compared to the average of 51±2 nm in FD. The difference in mean particle concentrations between the two periods was not significant, only 800 (#/cm3). Log-normal curve fitting to these mean distributions showed particle number concentrations for nucleation, Aitken, and accumulation modes of (4.4, 4.6, 2.5) x 103 #/cm3 in DS and (3.8, 4.1, 0.2) x 103 #/cm3 in FD. Regarding the geometric mean diameter (Dpg), values were (18.6 nm, 144 nm, and 53 nm) in DS and (19.8 nm, 55.6 nm, and 139 nm) in FD for the three modes. A possible explanation for this is the reduction in the number of vehicles in circulation during weekends, which is a source of precursor gases for particle formation and growth. Concerning particle concentration as a function of wind direction and intensity (PNCWD), results showed variations possibly related to emission sources and local atmospheric processes, as different configurations were observed during periods of intense vehicular traffic or under stable planetary boundary layer conditions, wind intensity, and orientation. The application of the K-means clustering algorithm to PNSDs revealed the presence of seven distinct clusters, each characterized by variations in predominant modes and their frequencies throughout the diurnal cycle. These results suggest potential influences from various sources, as well as dynamic and physico- chemical processes. In a subsequent stage, we plan to refine and expand our analysis, conducting a more detailed investigation of each of these clusters and establishing connections with the meteorological variables and aforementioned atmospheric constituents. Additionally, using the Mask R-CNN algorithm, which incorporates machine learning techniques, we aim to complement this analysis by identifying new particle formation events and obtaining particle growth rates and coagulation sink

Keywords: Air Pollution, Ultrafine Aerosols, Cluster Analysis, Particle Number and Size Distribution