The electrical gain of the measurement cell of all devices was measured by naneos under controlled conditions (Gain _lab ). Deviations from this Gain _lab may cause deviations in LDSA not related to the actual LDSA. As a result, a correction using Eq. 1 was applied to all datasets. Some devices developed some irregularities in the measured gain (Gain _measured ) leading to single spikes. These spikes in the Gain _measured were smoothed out, using the average between the gain before and after the spike, before conducting the gain correction. L D S A g a i n _ c o r r e c t e d = L D S A m e a s u r e d × G a i n l a b G a i n m e a s u r e d (1)

An electrical fan drove the flow rate through the tested devices. The fan’s performance was monitored as the fan’s current, and a flow sensor measured the flow. Unfortunately, the flow sensor became clogged during the long campaign in some devices. The true flow through the instrument was calculated post-campaign using a flow correction method from naneos, though the flow does not strongly affect the device’s signal [45].

The final clean-up involved removing unreliable data. Data were dropped where the difference between the measured data and the corrected data exceeded 30%. Apart from Reckenholz (4%), Schimmelstrasse (0.3%) and Kloten Feld (<0.1%), data loss was below 0.006% for the 2021–2022 campaign (Supplementary Table S3). In addition, due to weather damage to the Reckenholz device, unreliable data in July and August and from November onward were removed from further analysis; this was the only device damaged during the campaign. Where data were below the LOD of 2 μm²/cm³ [28], they were assigned the LOD. The percentage of data below the LOD ranged from 0.8% at Rümlang (rural) to 0.1% at Rosengartenstrasse (urban/roadside). Those greater than the upper limit of 20,000 μm²/cm³ [28], which were two values at Stampfenbachstrasse, were assigned the upper limit. For further data analyses, we used hourly averages of LDSA for agreement with the time resolution of other datasets. Adjustments to LOD and the maximum limit changed the means for the overall sampling period by −0.1–1.8 μm²/cm³ and the medians by −0.2–1.5 μm²/cm³ across all stations.

Data processing and analyses were conducted using Python version 3.7. For statistical analyses, we primarily used Scipy, a python package for scientific computing.

a. We used boxplots to visually represent the statistical distribution of LDSA at the different stations for the duration of the deployment.

Figure 2 presents LDSA pollution roses at the different stations, showing LDSA (colourmap) as a function of wind speed and direction. High concentrations occurred at low wind speeds at stations close to intense local sources of pollution. Wind speed between 4 and 7 m/s was generally associated with lower concentrations; higher speed blowing over high pollution sources was associated with high LDSA. This phenomenon is especially visible at Balsberg; high wind speeds from the west transported pollutants to the station.

With measurements starting in February 2021, LDSA was measured at most stations through all four seasons. 2021 had an unusually cool and wet summer and a relatively dry fall [Supplementary Figures S3, S4 (bottom right)]. Seasonal LDSA distributions in Supplementary Figure S4 reflect this seasonal weather pattern, showing highest median in the fall at most stations, dominated by high night-time concentrations (Supplementary Figure S5). There were however often more outliers in the winter at most stations (Supplementary Figure S4).

Brütten, the remote station, had the lowest LDSA. The range of median background LDSA at other stations agreed with that of total LDSA at Brütten, though the means differed (α = 0.05, Supplementary Table S4). Background LDSA was influenced by the location; while rural night-time background LDSA remained below 20 (±2) µm²/cm³ in all seasons, other locations sometimes recorded higher LDSA. Such high background was pronounced in the winter and fall, when there may be less sufficient atmospheric dilution to restore concentrations to lower background levels.

By the roadside stations (Rosengartenstrasse and Balsberg), LDSA peaked with the morning traffic rush-hour. In the warmer seasons, LDSA decreased through the afternoon/evening rush-hour, attributed to high atmospheric dilution as has been observed in other studies [39]. In the colder seasons, LDSA remained high, with some reflection of the afternoon/evening peak in the winter. At Balsberg, winter night-time LDSA decreased after the traffic peak. At Rosengartenstrasse, LDSA was influenced by the surrounding residential area at night, staying high after the traffic peak in the winter and fall, especially on weekends [10, 30, 39, 43].

Higher median pollutant concentrations at Rosengartengstrasse (except for LDSA and wintertime PM10) than at Balsberg may be related to the traffic flow especially during peak hours [55] and residential combustion. At Rosengartenstrasse, the speed limit is 50 km/h, and traffic is more prone to being congested than on the motorway next to the Balsberg station (traffic congestion levels estimated by Google Maps [56]). The speed limit on the motorway next to the Balsberg station is regulated between 80 and 100 km/h based on the traffic volume for better traffic flow [57]. Higher frequency of acceleration and deceleration in traffic congestion has been shown to increase vehicle emissions [55]. Such pollution creates exposure hotspots for residents close to Rosengartenstrasse.

We observed seasonal differences in the correlation between LDSA and co-pollutants. High correlations between NOx and LDSA indicate traffic influence on LDSA and, like other studies [10, 22, 39, 42, 43], was highest in the winter. We observed similar correlations to other studies between NOx and LDSA (Table 1). Since the influence of wind direction was not considered, the traffic counts considered here may not reflect the true impact of traffic, hence, the low correlations with traffic. We also observed similar PM_mass correlations with LDSA to other studies (Table 1). PM10 and PM2.5 had lower correlations with LDSA due to the dominance of particles less than 100 nm in traffic PM emissions [16, 17]. The PM_mass correlations also differed with seasons between both stations; highest correlations were in the winter at Balsberg, but lowest in the winter at Rosengartenstrasse. These differences may be due to residential heating in the cooler seasons at Rosengartenstrasse [30].

The Kloten station is influenced by both road and air traffic, especially when wind blows from the west- northwest direction; this is visible in the pollution rose at Kloten Feld (Figures 2, 4C). The influence of road traffic is evident in the morning peak in LDSA; similar roads to those at Balsberg contribute from the west-northwest to LDSA at Kloten. The landing path of the Zurich airport’s “east approach concept,” which is between 21:00 and 23:30 on weekdays and 20:00 and 23:30 on weekends [54], is visible on the pollution rose as a trail of high LDSA in the southeast-northwest direction and as night-time peaks in LDSA (Figure 4D) when traffic is at its lowest (Balsberg, Figure 3). The earlier and broad (weekend) fall and winter peaks in night-time LDSA may be due to additional residential combustion (not present in the spring and summer); and the later peak (winter weekdays) reflects the air traffic influx. The seasonal differences in air travel are also visible (Figure 4D): summer night-time LDSA were as high or higher than winter and fall LDSA with higher summer flight movements [58]. This air traffic influence on ground air quality is similar to previous observations with PNC, showing airport influence up to 5 km from the airport and 300 m above ground [35, 59]. The question about the health impact on residents along this flight path remains open.

Ambient particle size distribution (PSD) changes with distance from the sources of emission, with sub-100 nm dominant close to combustion sources [16, 17, 19, 60, 61]. The presence of these particles is not reflected in PM mass metrics, but in PNC and LDSA. We observed the influence of the nature of the sampling location on different PM metrics at Kaserne and Stampfenbachstrasse, where PNC and PM_mass are also monitored. Like in other case studies, winter and fall had the highest pollutant levels.

Kaserne is a partially closed square with food trucks and is in the tourist hub of Zurich. The monitoring station here is about 50 m from the food trucks. The orientation of the buildings around the square may isolate the square from the bulk of emissions from surrounding roads and may also allow for retention and recirculation within the square [62]. Being a region for social activities, we observed some morning increase in emissions with increase in activities in the city, and especially, an increase in the evening with social and nightlife activities. Low atmospheric dilution probably drove the night-time high concentrations in the fall and winter.

At Stampfenbachstrasse, the LDSA device was located next to a pizzeria’s kitchen window. Here, the kitchen’s morning opening hours are reflected in the second morning LDSA peak; the start of the evening peak also coincided with the evening operating hours. A previous study has observed high PM emissions, including LDSA, in pizzerias [63]. Our data therefore also show that high-combustion eateries may create local outdoor hotspots with potential influence on indoor air quality in neighbouring residences.

Regarding other PM metrics, there was good rank correlation (ρ) between LDSA and PNC. The combination of some distance from the source of emission and recirculation at Kaserne may have allowed ageing of particles, including particle growth [62]. Such ageing allowed for good correlations between LDSA and PM_mass, and moderate correlation between PNC and PM_mass. PSD measurements will better inform the suggested particle ageing.

Stampfenbachstrasse, on the contrary, does not have the possibility for particle retention/ageing, leading to lower correlations between LDSA and PM_mass, and PNC and PM_mass. PM_mass measurements were about 10 m from the pizzeria. The correlation between LDSA and PNC was good, though lower than at Kaserne, as LDSA still misses particles less than 20 nm [28, 29, 41]. There were also differences in the diurnal patterns of the different metrics; both LDSA and PNC clearly captured the kitchen’s schedule unlike the mass metrics. These data show that mass metrics alone here may miss the high exposure of nearby residents.