Particle size distributions and PM concentrations are monitored by a Palas Fidas 200 (Palas GmbH, Karlsruhe, Germany), an aerosol optical spectrometer designed for regulatory purposes and referenced in various studies . The instrument determines PSDs in 63 logarithmically spaced bins across a particle diameter range of 0.18–18 µm using the principle of aerosol light scattering. The implications of the minimum diameter detected by the instrument are discussed in Sect. . Based on the PSDs, the Palas Fidas 200 also retrieves PM₁, PM_2.5, PM₄, and PM₁₀ concentrations, with equivalence certification by TÜV for PM_2.5 and PM₁₀. Details on the built-in proprietary retrieval algorithm, PM_ENVIRO_0011, can be found in the instrument manual . Some information is reported in Sect. S2 together with a more exhaustive description of the instrument operating principles, the sampling drying system, and an approximate quantification of the measurement uncertainty.

For this study, which primarily focuses on PM₁₀ mass concentrations, the number size distributions from the Palas Fidas 200 are further converted to volume size distributions (VSDs) under the assumption of spherical particle shape. Additionally, the resulting VSD is adjusted using the typical US-EPA efficiency curve for standardised sampling inlets as defined by EN 481 (with a 50 % cut-off at 10 µm) and detailed in the Palas Fidas 200 manual . These conversion steps are optional, as the PMF optimisation metric (introduced in Sect. , Eq. ) includes a normalisation based on the estimated uncertainty and is thus insensitive to the type of distribution provided. Nevertheless, we anticipate here that an important step is to select a PM₁₀ efficiency curve that includes diameter bins larger than 10 µm in the analysis, without cutting the VSD at 10 µm, in order to improve the separation between desert dust and local resuspension contributions (Sects.  and ).

The dual-spot Aerosol Magee Scientific AE33 aethalometer continuously determines the light absorption coefficient, babs(λ), of particles deposited on a tape, at seven wavelengths spanning from ultraviolet (UV, 370 nm) to infrared (IR, 950 nm). The spectral dependence of babs(λ) is commonly parameterised by an exponential function, defined by an Ångström Absorption Exponent AAE;. The AAE facilitates the discrimination of carbonaceous aerosol species such as black/brown carbon BC/BrC; e.g.,. However, the optical properties of carbonaceous aerosols show large variability depending on particle size and chemical composition, and thus on the measurement location and period , therefore caution is advised when using predefined (a priori) AAE values in the optical source apportionment.

The babs value at 880 or 950 nm can be converted to eBC mass concentration, expressed in ng m⁻³, using a mass absorption cross-section (MAC) coefficient. This conversion is actually operated by the instrument at all wavelengths, based on a set of manufacturer-defined coefficients that scale inversely with wavelength, thus accounting for the theoretical spectral dependence of black carbon absorption and leading to the “nominal eBC” (NeBC), as defined by . Employing NeBC(λ) “mass concentrations” calculated as such instead of babs(λ) allows for a simpler computation of spectral absorption differences. For instance, positive differences between NeBC mass concentrations in the UV and IR spectral ranges are indicators of UV-absorbing compounds, such as those associated with biomass burning. This quantity, often referred to as “Delta-C” in the scientific literature , is included as a variable in our PMF analysis (Sect. ): 1Delta-C=NeBC(370nm)-NeBC(880nm)

The choice of the upper limit (880 or 950 nm) is not critical. Conversely, we selected the lower limit based on both the correlation index between levoglucosan concentrations and aethalometer measurements at 370 nm, and the temporal patterns of Delta-C (further details are provided in Sect. S3). Although recent research demonstrates that using instrument- and site-specific parameters , or harmonised coefficients , leads to more accurate determination of the absorption coefficients, we employ nominal values since the source apportionment results within the examined time frame are influenced more by short-term temporal and spectral variations in aerosol light absorption than by the absolute accuracy of babs values.

In any case, it is worth highlighting that our RASPBERRY algorithm directly incorporates all 7-wavelength measurements from the aethalometer, in contrast to other studies that rely on a priori assumptions of the AAE for the apportionment of fossil fuel and biomass burning sources e.g.,. Additional details regarding the aethalometer instrumental setup and maintenance can be found in Sect. S4.

Common EU regulated air pollutants are monitored at the Aosta–Downtown station. In this study, we employ NO_x hourly concentrations determined with Teledyne API200E and Horiba APNA370 chemiluminescent analysers as proxy for combustion processes in a posteriori correlation analysis. We opted not to include gases in the source apportionment, unlike some previous studies , as gases might undergo different processes compared to PM₁₀ and their photochemical behaviour can be highly dependent on meteorological conditions. More details on this aspect are provided in Sect. S5.

The chemical PMF setup closely follows the methodology described by . An important difference in this study is the use of PM₁₀ concentrations from the Palas Fidas 200 as the total variable for the chemical PMF, replacing the data from a previously co-located Opsis SM200 beta attenuation monitor (no longer available in the period under investigation). This change also ensures consistency with the physical PMF, based on hourly PM₁₀ concentrations and size distributions from the same Palas Fidas 200 instrument. Analytical error fractions and detection limits reported by ARPA chemical laboratory are applied using an equation-based approach to estimate total uncertainties for each sample/species combination. As outlined in Sect. , two separate chemical PMF analyses are performed due to the alternating sampling schedule and differing chemical characterisations. Further information on the configuration of the chemical PMF is reported in Sect. S6.

The physical PMF analysis incorporates hourly averages of volume size distributions derived from the Palas Fidas 200 and multiwavelength optical absorption data obtained from the aethalometer. For the size-related measurements, we employ 63 VSD bins dVdlog⁡10D with mid-point diameters (Dp) ranging from 0.2 to 17.17 µm, without any channel grouping (further details are provided in Sect. S7). For the optical absorption part, we include multiwavelength measurements from the aethalometer, specifically NeBC at six wavelengths from 470 to 950 nm and Delta-C as defined in Eq. (). NeBC(370 nm) is excluded to avoid collinearity with Delta-C. The PM₁₀ concentration measured by the Palas Fidas 200 is retained as the “total variable” .

It is worth emphasising that dimensional and optical absorption information is combined in this study within a single PMF analysis. This approach differs from the methodology presented, for example, in , where two separate PMFs are conducted sequentially – one based on chemical information and the other on size distribution data – to determine the size distributions associated with specific chemical characteristics. Our approach also differs from those of and , who performed a post-hoc multi-linear regression between particle number measured in different size bins and source contributions from a prior chemical/elemental PMF. Conversely, our method integrates dimensional and light absorption properties into a unified PMF, both contributing to shaping the final solution and stabilising it by reducing rotational ambiguity in the profiles .

A critical step for achieving successful source apportionment is the definition of uncertainties to be inputted into the PMF . In this study, we adopt the same uncertainty framework described by , which is often employed in PMFs based on particle size/number distributions. The measurement uncertainty for each data point is modelled as: 4σij=AαNij+N‾j where Nij is the size distribution for sample i and size channel j, N‾j is the time average in channel j, and A and α are free parameters. The overall effective uncertainty is subsequently defined as: 5sij=σij+C3Nij where C3 is an additional parameter to be selected. For this study, the formulae utilise particle volume distributions (rather than number distributions, N) for the dimensional data, and NeBC mass concentrations at aethalometer wavelengths for the optical absorption data.

In previous works, the uncertainty configuration is typically addressed pragmatically through trial-and-error procedures or iterative approaches that explore combinations of parameter values to optimise a given metric . Moreover, when different quantities are combined within a single PMF analysis, their residuals must be appropriately weighted in Q to ensure that each quantity exerts a balanced influence on the final solution (i.e., through the total contribution of their scaled residuals to Q). Throughout this procedure, we deliberately avoided “tuning” the physical PMF results to reproduce those obtained from the chemical PMF, thereby preserving the independence of the two datasets. The full procedure is described in Sect. S7, where an objective and reproducible workflow is presented for interested readers. The final coefficients A, α, and C3 adopted for the PMF analysis are reported in Table S1. The field Extra Modelling Uncertainty of the EPA PMF was left unchanged (0 %). These values refer to the input uncertainties used in the training (PMF) phase of RASPBERRY. A more comprehensive overview of the method limitations and of the overall uncertainties associated with the retrievals is provided in Sects.  and .

Seasonal splitting was attempted, but without satisfactory results. This is discussed in Sect.  (and Sect. S8). No normalisation for the dilution effect was performed a priori on the input dataset, since under many conditions the wind contributes to aerosol advection rather than dilution at the study site . An alternative approach, based on a posteriori meteorological normalisation and applied to the source-apportioned results e.g., will be described in a forthcoming publication.

The first two factors are assigned to road traffic and residential biomass burning emissions, respectively. Indeed, they both exhibit strong light absorption in their profiles (Fig. a and c), with NeBC contributions significantly different from zero as evident from the displacement interval of the perturbed variables (constrained DISP test, coloured areas). Traffic emissions (Fig. a) show an average NeBC mass concentration of approximately 700 ng m⁻³ and a Delta-C of zero, indicating small spectral variation from IR to UV wavelengths, as expected from BC-dominated particles. This factor accounts for about half of the total NeBC at 370 nm and even more at longer wavelengths. Biomass burning (Fig. c) has a slightly lower contribution to NeBC (400–600 ng m⁻³) and a Delta-C accounting for nearly 100 % of its total value, denoting increased absorption in UV wavelengths due to light-absorbing OC (BrC). Ex-post AAEs calculated using absorption coefficients at all wavelengths are 1.1 for road traffic and 1.8 for residential biomass burning. These values are consistent with established AAE ranges of 0.9–1.1 for liquid (fossil) fuel combustion and 1.7–2.2 for biomass burning both at the surface e.g., and for the total atmospheric column .

Regarding particle size, a detailed comparison between our results for each source factor and previous literature is provided in Sect. S14. Here, we restrict the discussion to the main considerations. First, the size distributions of both factors (Fig. b and d) exhibit multimodal structures, indicating composite source contributions . A common feature, however, is their increase towards the lower limit of the diameter scale, suggesting potentially higher VSD values for particles smaller than 180 nm, which is beyond the lower OPC detection limit. We acknowledge the significant limitations in identifying traffic-related particles based on OPC-derived VSDs. Notably, the largest number contribution from “fresh” traffic exhaust particles typically lies within the Aitken mode tens of nanometres, e.g., which is not captured by the OPC. Nevertheless, in our case, aethalometer measurements – characterised by high absorption coefficients and an AAE close to 1, the commonly accepted theoretical value for black carbon – play a fundamental role in attributing this factor to traffic exhaust emissions, particularly to an “aged traffic” component that has shifted to larger particle sizes . This interpretation is further supported by the associated temporal patterns discussed in the following paragraphs.

Some particles within this factor, especially those with diameters greater than 1 µm, may additionally originate from non-exhaust emissions, and notably tyre and brake wear , road surface abrasion and dust resuspension . Only a small fraction of particles > 5 µm is visible in this factor, as these latter are likely better represented by factor 6 (Sect. ). The differing size fractions of exhaust and non-exhaust particles, along with their distinct atmospheric dispersion behaviours and sensitivity to weather and pavement conditions, can lead to partial decorrelation between these particle types. This effect is particularly evident in sub-daily measurements, where temporal patterns of the two fractions may shift relative to each other (see further results on this aspect in Sect. ). It is also worth noting that even source apportionments based on aerosol chemical properties face limitations in attributing all coarse particles from non-exhaust emissions to traffic. However, in that case, the temporal decorrelation between exhaust and non-exhaust particles may be partially alleviated by use of daily averaged data.

As a final remark on the size distributions, the dQmax⁡=4 region of both combustion-related factors (Fig. b and d) is the largest among all PMF profiles in relative terms. This is also confirmed by the relatively large error bars obtained with RASPBERRY+EVLS in Fig. S34 for the same sources. A likely explanation for this behaviour is the greater uncertainty in the VSDs of these factors.

The temporal patterns (Fig. a) reveal similarities and differences. Both factors show two maxima in their daily cycles, with peaks in the morning/evening and minimum in the middle of the day. This reflects the daily evolution of the mixing layer in the valley and the emission cycles, with traffic peaking during rush hours and biomass burning, associated with operation of residential heating systems, peaking approximately 3 h later. The differing behaviour of the two emission sources becomes even more apparent when their relative contributions to PM₁₀, rather than absolute values, are considered (Figs. S13a and S20–S21). Notably, biomass burning accounts for 30 %–40 % of nighttime PM₁₀ during the winter months. Seasonally, traffic emissions contribute quite consistently to PM₁₀, however, during the cold season, the morning and late afternoon peaks become more pronounced (Fig. S14), likely due to the reduced mixing height. A slight increase in PM₁₀ from traffic during summer and a marked rise in December are observed (Fig. a), probably due to tourism . Indeed, in winter Aosta is a prominent destination for skiers frequenting nearby snowfields, particularly during the winter holiday season. On the other hand, biomass burning is confined to winter.

Weekly trends further distinguish the two, with road traffic emissions showing a pronounced weekend effect, in contrast to biomass burning. In particular, the Sunday morning peak of traffic emissions is remarkably damped compared to the other days of the week (Fig. S14). This difference is confirmed by the Kruskal-Wallis test e.g.,, used to check whether daily mean PM₁₀ contributions are similar on weekdays and weekends (null hypothesis). The resulting p-values are 1 × 10⁻¹⁷ for traffic emissions, i.e. weekdays/weekend differences are statistically significant at the 5 % level, and 0.89 for biomass burning, i.e. no significant differences. Despite this, the biomass burning morning peak exhibits a slight weekday/weekend difference, potentially indicating a weak interference from traffic emissions. A similar behaviour was identified by , who suggested that the AAE of traffic emissions may vary throughout the day, with larger values – mimicking that attributed to biomass burning – for fresh emissions. This effect is expected to be more pronounced in low BC concentration scenarios, such as in our study. If this is the case, the observed behaviour is intrinsic to any aethalometer source apportionment model based on only two factors. An additional explanation proposed by the same authors involves the rapid formation of secondary organic aerosol from the ageing of traffic also, which again leads to an increase in AAE.

The overall contributions of these two factors to the total PM₁₀ in the period 2020–2024 are 1.6 µg m⁻³ (9 %) for traffic and 1.8 µg m⁻³ (10 %) for biomass burning. These relatively low fractions reflect the generally unpolluted nature of the site, with weak local emission sources. However, it should be noted that these values represent annual averages, whereas wintertime concentrations can be significantly higher (e.g., Fig. S8).

As confirmation of the correct attribution of these factors, three long-term statistical considerations are provided. First, the sum of the two factor contributions, representing combustion-related PM₁₀, is correlated with the NO_x concentration measured at the same station over the 5-year period 2020–2024. Figure S11 shows that, despite the different physical states of the pollutants (particles and gases), their relationship is linear, with a Pearson correlation coefficient of 0.93 (R2 = 0.87). Second, the average daily cycle of PM₁₀ concentrations attributed to traffic at Aosta–Downtown is compared with vehicle counts recorded simultaneously 500 m to the south during a measurement campaign conducted in 2020–2021 (231 measurement days distributed throughout the 2 years). Although rigorous and sophisticated methods exist to disentangle the effects of emissions and meteorology e.g.,, which will be the focus of a separate study, Fig. S12 confirms that the two quantities are well correlated, exhibiting similar hourly and weekly patterns. Finally, the average daily cycle of traffic-related contributions, as determined by RASPBERRY, is compared over the period 9 March–4 May across different years (Fig. ). This period was selected as it corresponds, in 2020, to the strictest phase of the COVID-19 “lockdown”. The figure qualitatively shows that the reduction in traffic emissions due to the containment measures had a marked impact on air quality in Aosta. While the overall PM₁₀ concentrations did not vary substantially – partly due to the influence of meteorological conditions – the effect of the lockdown on the composition of the aerosol mixture is clearly discernible. This finding updates the results of , providing unprecedented high-time-resolution insights into the “lockdown effect”.

Based on prior literature and studies conducted in the region , factors 3 and 4 are attributed to secondary particles in condensation and droplet modes . Indeed, secondary aerosols are well-known contributors to submicron particles in the accumulation mode (; ), with many studies identifying two sub-modes at distinct diameters . In particular, our results indicate that their relative contributions to the volume size distribution peak at 250 nm for factor 3 and 500 nm for factor 4 (Fig. f and h). These modes have been associated with different formation mechanisms: gas-phase processes, resulting in smaller particles the so-called “condensation” mode, e.g., and mixed-phase processes, yielding larger particles (the so-called “droplet” mode). This attribution to secondary particles is also consistent with their weak light absorption (Fig. e and g). In particular, secondary inorganic aerosols, rich in sulfate and nitrate, are generally characterised by low absorption coefficients . Nevertheless, the droplet mode exhibits greater variability in absorption coefficients and Delta-C, which may, for instance, indicate presence of organic compounds (e.g., formation of organic nitrates).

The diurnal temporal patterns are remarkably similar, with a primary maximum in the late afternoon, a secondary peak in the morning, and a minimum just after midday (Fig. b). The observed concentration daily maxima can be attributed to two processes. First, and likely predominant, is the transport of polluted air masses, enriched in secondary particles, from the Po Basin to the Alps. This transport occurs regularly in the Aosta Valley during sunny days with weak synoptic circulation, accounting for approximately 50 % of the days annually and peaking in the afternoon. During such events, surface concentrations of fine secondary particles may be further amplified by reduced vertical mixing towards the end of the day. The strongest and prolonged transport episodes, leading to accumulation of particles, are clearly visible as peaks (Fig. S9c and d), which is confirmed by remote sensing techniques (some examples are provided in Sect. ). A second, yet unexplored, reason for the afternoon increase could be the local formation of secondary particles after sunset, facilitated by rising relative humidity, favourable meteorological conditions such as atmospheric stability, and presence of local or advected particles promoting secondary formation through heterogeneous reactions. The subsequent nocturnal decrease in concentrations is likely driven by drainage winds in the valley, as also observed for the traffic-related component. The secondary morning maximum could arise from several mechanisms: (i) local secondary particle formation linked to emissions, such as traffic during rush hours; (ii) entrainment of secondary particles from the nighttime residual layer, acting as a reservoir overnight ; or (iii) the initial stages of a progressive accumulation of secondary particles throughout the day, interrupted by a sudden concentration drop at midday due to enhanced vertical mixing . Determining the dominant process needs further investigation. Anyway, the absence of a weekend effect for both factors, as indicated by p-values from the Kruskal-Wallis test well above 0.05 (0.69 for the condensation mode and 0.70 for the droplet mode), suggests that local anthropogenic emissions of aerosol precursors play a minor role. Instead, concentrations appear to be predominantly influenced by regional-scale atmospheric circulation patterns, accumulation processes, and meteorological/thermodynamic conditions.

Seasonally, the contributions to PM₁₀ by condensation and droplet modes are comparable in winter, while from April to September the condensation mode is dominant, accounting for up to 40 %–50 % on an hourly basis during the night (Figs. S22–S23). The decrease in the concentrations of the droplet mode factor during summer (Fig. b) is a well-documented phenomenon attributed to less favourable formation conditions and the partitioning of compounds such as nitrate ammonium towards the gas phase under warmer conditions e.g.,. The seasonal modulation of the condensation mode is less clear, with a minimum in April–May followed by a rapid increase and a secondary peak in July. Very interestingly, the same distinct minimum in the month of May has been found in Milan in ammonium sulfate concentrations by . The seasonal behaviour of the condensation mode factor may be linked to (i) varying mesoscale or synoptic circulation patterns (e.g., the transport of sulfates from other European countries) or (ii) enhanced photochemical formation processes in summer.

The overall contributions of these factors to the total PM₁₀ in the period 2020–2024 are 4.1 µg m⁻³ (23 %) for the condensation mode and 2.8 µg m⁻³ (16 %) for the droplet mode. These values closely correspond to those determined through chemical analyses by , who estimated the contribution of secondary aerosols (sum of sulfate- and nitrate-rich factors) in Aosta–Downtown to be in the range of 30 %–40 %.

Factors 5 and 6 represent coarse, predominantly non-light-absorbing, particles, as shown in Fig. i–l. We attribute them to the long-range transport of mineral dust from desert areas and resuspension of soil particles of more local origin, respectively.

For desert dust, this interpretation is mainly supported by results obtained using independent remote sensing techniques, the analysis of back-trajectories and the CAMS Ensemble model, as discussed further below, as well as the characteristic peak-like, impulsive time series of this factor Fig. S9e with an isolated average increase in July and a minimum in December, shown in Fig. c. Another indicative feature of long-range transport is the weak dependence of the PM₁₀ contribution on the time of day and the day of the week, likely due to the “random” arrival times of these air masses at the site via long-range circulation (Fig. c). The small but statistically significant decrease in weekend concentrations (p-value = 2×10-5 from the Kruskal-Wallis test) may be attributed to reduced resuspension of deposited dust by vehicular traffic , or to contributions from other local sources. The size profile, peaking at approximately 5 µm, with the maximum contribution to VSD variance occurring over a relatively broad range of diameters centred around 2 µm, is consistent with existing scientific literature on desert dust transported towards Europe. For example, report that long-range transported dust in continental Europe, identified using lidar and satellite observations, typically has diameters ranging from 0.7 to 3 µm, whereas locally resuspended coarse particles exhibit larger diameters. Comparable values are reported in other studies . note that desert dust in Genoa, Italy, is characterised by a broad range of diameters extending from 0.5 to over 4 µm. In their review, state that clay-like dust typically has a size of ∼ 2 µm, while silt-like dust is larger, at around 5 µm. Finally, demonstrate that desert dust accounts for the majority of variance in the 2.5–3 µm range, consistent with our findings, also considering that their measurements were conducted in Lecce, further south in Italy.

Regarding optical properties, this dust factor does not exhibit significant light absorption, in contrast to previous findings . However, unlike remote, pristine sites , the absorption by dust is often masked in regions heavily influenced by other light-absorbing aerosols, such as traffic and biomass burning emissions , due to its lower mass absorption efficiency . Moreover, the absorption characteristics of desert dust can vary significantly depending on its source region . Hence, not all studies identify a dust factor with light-absorbing properties, even at southern European stations e.g.,. However, despite the uncertainty in NeBC encompassing the zero line in Fig. i, we note that the optical profile of factor 5 increases at wavelengths shorter than 600 nm, consistent with expectations for dust . For our study, the estimated AAE of approximately 3 agrees well with the upper limit of AAE values for dust detected at AERONET sites .

As anticipated, ancillary information from remote sensing techniques and models confirms the correct attribution of this factor. Figure a presents the coarse-mode AOD retrieved from the sun photometer using the algorithm by (Sect. ). The year 2021 is chosen as an example, as some of the strongest transport events in Europe occurred in that period . An arbitrary minimum threshold of 0.15 on coarse-mode AOD is set to highlight the most indicative episodes in the plot. Furthermore, based on the analysis of vertical profiles from the ALCs in Aosta–Saint-Christophe, dust layers that remain primarily aloft (light blue bands in Fig. b), detected by the sun photometer but not by the in-situ surface instruments, are discriminated from those that ultimately enter the mixing layer and reach the ground (yellow bands in Fig. b). Representative examples are shown using lidar diagrams in Fig. S31a–d. In the latter cases, the contribution of the desert dust factor in RASPBERRY increases markedly, whereas for the former, the increase is negligible. This confirms that factor 5 serves as an effective proxy for the presence of desert dust. Slight delays are occasionally observed between detection by remote sensing instruments and peaks in the source apportionment. This may be attributed to various effects: (i) absence of photometer measurements in cloudy days; (ii) time required for the layer to descend after being detected in the column or, in some cases, (iii) time needed for the dust to be advected horizontally to the measuring station (and there accumulated) after entering the atmosphere elsewhere (e.g., the Po Basin). The third plot (Fig. c) presents the results of the concentration-weighted trajectory analysis CWT; e.g.,, using the HYSPLIT model and the dust factor contribution as the weighting variable (more details in Sect. S16). The figure clearly shows that the most likely source region for the particles attributed to factor 5 is northwestern Africa.

In addition, a comparison of the daily average concentration of desert dust in surface PM₁₀ from RASPBERRY and the CAMS Ensemble Validated Reanalysis VRA; for the year 2022 is presented in Fig. d. This year was selected as it corresponds to the study period currently under evaluation within the CAMS–National Collaboration Programme–Italy (CAMS2_72IT_bis). The agreement between the two datasets is notable, both in terms of the timing of dust events (x-axis) and the absolute concentrations (y-axis). The most pronounced differences in concentration occur during the June 2022 event, with RASPBERRY showing peak values approximately twice as high as those from CAMS Ensemble VRA. In this context, both overestimation and underestimation by CAMS relative to surface in situ observations have been documented in the literature, depending on spatial and temporal variability as well as on ancillary conditions . Potential systematic biases may also arise from optical particle counter (OPC) artefacts under dust conditions. However, correcting for these effects would require detailed knowledge of the aerosol refractive index, and thus of its chemical composition, and morphology at high temporal resolution . Finally, a persistent non-zero background is evident in the RASPBERRY dataset throughout the year, as discussed further below.

The attribution of factor 6 to local coarse particles is supported by the pronounced weekend effect (p-value of 2×10-26), which indicates an anthropogenic origin (Fig. c). Moreover, this factor is shifted to larger diameters, with a peak of the absolute contribution between 5 and 6 µm, and its variance contribution extending to the upper limit of diameters detectable by the Palas Fidas 200. Such large particles are unlikely to travel long distances. These particles may include: (i) crustal materials resuspended from the road pavements by vehicular traffic (Sect. ) or steelwork slag dust, both characterised by high calcium and magnesium content; (ii) sodium chloride particles originating from road salting used in winter as a de-icing agent; and (iii) pollen and other primary biogenic aerosols. The influence of road salting likely explains the observed increase in coarse particle concentrations during December, January, and February (Fig. c, see also the case study presented in Sect. ). The average diurnal pattern, with a peak around midday, resembles the evolution of the convection-driven aerosol layer height, as determined in previous studies at the same site using automated lidar-ceilometers , highlighting the role of local-scale mixing. At a closer examination (Figs. S18–S19), a single daily peak is observed at midday during January and December. Conversely, in other months, two distinct peaks emerge. This phenomenon can be attributed to the influence of local atmospheric (valley) circulation, where winds develop around midday and, if strong enough, erode the temperature inversion, dispersing particles into a larger volume of air. The two peaks are unlikely to originate from traffic emissions during rush hours, as their separation remains evident even during the typical summer holiday months. Figure S32 illustrates this behaviour on a representative summer day (18 July 2024). Vertical aerosol profile measurements from polarisation-sensitive ALCs, such as the CL61 (Sect. ), provide valuable information in this context, as high depolarisation ratio values are indicative of the presence of irregularly shaped particles in the atmosphere. Notably, the morning increase and evening decrease in surface concentrations visible in Fig. S32 correspond well with the diurnal evolution of the depolarisation profiles. However, around midday, when the particles are transported and mixed into a larger air volume at altitudes exceeding 1.2 km, surface concentrations decrease. Another interesting feature is that in December the daily peak in coarse particle contribution is shifted towards the afternoon, whereas in January it is more centred around noon or slightly shifted towards the morning. This fact could indicate a different daily pattern in car traffic during the winter holiday period, as discussed in Sect. .

The overall average contributions of these two factors to the total PM₁₀ in the period 2020–2024 are 3.6 µg m⁻³ (21 %) for desert dust and 3.7 µg m⁻³ (21 %) for local resuspension of coarser particles, together accounting for more than 40 % of the total PM₁₀. Such a large percentage contribution is justified by the substantial volume carried by these coarse particles and the relatively low contributions from other local sources at this lightly polluted measurement site. In particular, according to RASPBERRY desert dust estimates, 22 out of the 36 PM₁₀ daily exceedances recorded in Aosta–Downtown during the 5-year study period (as defined by the new 2024 AAQD; 16 out of the 26 under the current 2008 AAQD) could in fact be excluded from the count due to the contribution of natural sources. At the same time, we acknowledge that a slight overestimation of the desert dust contribution by RASPBERRY may be possible. Indeed, after removing peak events and data within ±12 h of such episodes, the residual baseline averages ∼ 2 µg m⁻³, a value compatible with PM measurement uncertainty. Notably, this baseline exhibits a distinct diurnal and weekly variability (including a morning maximum and weekend effect), as well as seasonal features (peaks in July and October), suggesting a contribution from the resuspension of fine crustal particles (of desert or local origin) driven by traffic and modulated by road surface moisture, and not captured by the CAMS ensemble. Despite these limitations, factor 5 remains highly effective for identifying desert dust events and the overall results remain qualitatively consistent with similar dynamics observed in other southern European regions.

The period from 27 to 31 December 2024 was characterised by high PM₁₀ concentrations across the city, as indicated by all urban quality monitoring stations. RASPBERRY attributes these elevated concentrations primarily to locally resuspended coarse particles, whose hourly contribution reaches the PM₁₀ daily average limit introduced by the 2024/2881/EC AQ directive (45 µg m⁻³). The temporal pattern of these contributions remains remarkably consistent over the days (Fig. S35), with maxima occurring between 16:00 and 17:00 CET. This winter holiday period was marked by dry conditions, the inactivity of the steel mill, and the absence of construction works within the city. In contrast, vehicular flow associated with winter tourism was significant, making it an ideal case study for assessing the impact of exhaust and non-exhaust emissions from car traffic.

Retrievals from the physical source apportionment for factor 1 (“traffic emissions”, primarily exhaust and brake/tyre abrasion, Sect. ) and factor 6 (“local dust resuspension”) are presented in Fig. . The first factor (Fig. a) exhibits a primary maximum in the morning (around 10:00–11:00 CET), with no corresponding peak in the sixth factor. The same morning peak is also clearly visible in nitric oxide (NO) concentrations (Fig. S36a), confirming its attribution to combustion (mainly traffic) emissions. The afternoon peak in factor 6 (Fig. b) is also discernible in factor 1 and NO concentrations, though less pronounced, suggesting a common traffic-related source. Notably, this coincides with the closing times of cable cars operating between Aosta and nearby ski resorts, a well-known rush hour in the southern part of the city.

Road salting is a strong candidate for the observed increase in coarse particles in the afternoon. Sodium chloride, commonly used as a de-icing agent in the region, can reach daily average concentrations exceeding 20 µg m⁻³ on some winter days, nearly doubling those of the crustal-related factor in chemical PMF . Since particle resuspension is strongly influenced by weather and pavement conditions, and in particular by surface moisture , the absence of a morning increase in coarse particle concentrations during winter may be attributed to humid or frozen road surfaces. As the day progresses and the sun rises over the mountain horizon around midday, relative humidity decreases (Fig. S36b), likely leading to drier road conditions. This short sunlit period is also reflected in the wind speed diagram (Fig. S36c), which shows a slight increase in wind speeds, though still very low and close to calm conditions (∼ 1 m s⁻¹), likely due to strong wintertime temperature inversions. While this weak atmospheric circulation is insufficient to resuspend particles from the road surface on its own, it may contribute to lifting particles already resuspended by vehicular traffic. As the sun sets behind the mountains, resuspended coarse particles settle back onto the surface, while the simultaneous afternoon traffic peak further contributes to the observed PM₁₀ increase.

Although optical particle counters (OPCs) may partly overestimate sodium chloride concentrations , this example highlights the significant (and unregulated, in the current AAQD) contribution of non-exhaust emissions, and notably road salting in mountainous regions during winter, to PM pollution . It is worth noting that these emissions are not expected to decline with the transition to electric vehicles. The example also underscores the fact that exhaust and non-exhaust traffic emissions cannot be fully captured by a single source apportionment factor.

Accidental fires are typical events requiring real-time and high-temporal-resolution air quality monitoring, as they demand immediate attention from municipal surveillance bodies, the public, and the media. Since residential biomass burning contributions are negligible during the warm season (Fig. a), such events are relatively easy to spot in summer data. One such incident occurred on the night of 22 August 2022, when a restaurant woodshed (used for a wood-fired oven) in the city center caught fire. Although the event lasted only a few hours, it was clearly identified by a peak in the biomass burning component (Fig. a), demonstrating the effectiveness of the “biomass burning” source apportionment factor as a specific marker for such occurrences. Due to the uncontrolled nature of the fire, the emitted aerosol spanned a broad size range, overlapping with other factors (Fig. b). Given the greater mass contribution of larger-mode factors, concentrations associated in this case with the “droplet mode” peaked at a value of 80 µg m⁻³, i.e. nearly the total PM₁₀ concentration, which may also reflect the presence of organic compounds within this factor.

The mesoscale circulation between the Po Plain and the Alps frequently transports aerosol-rich air masses to the measuring site , exacerbating air pollution and contributing to PM exceedances. An episode in March 2022 was especially noteworthy due to the co-occurrence of such secondary aerosol transport and Saharan dust.

Figure a illustrates the RASPBERRY retrievals on 17–18 March, showing desert dust contributions at the beginning and end of the period. The same dust event has been documented in other southern European countries . In contrast, the middle of the episode is dominated by an increase in droplet-mode particles, characteristic of wintertime secondary aerosols, with PM₁₀ contributions of up to 45 µg m⁻³. This is further supported by our chemical analyses, which indicate a daily averaged PM₁₀ contribution from the nitrate-rich factor identified by the chemical PMF larger than 25 µg m⁻³ on 18 March 2022 in Aosta–Downtown. The increase in droplet mode coincides with eastern surface winds exceeding 5 m s⁻¹, typical of the breeze circulation regime originating from the Po Basin. During the same period, secondary ammonium nitrate and possibly secondary organic aerosols were indeed detected in the Po Valley by and . Heterogeneous chemical interactions with mineral dust may have also played a role. A slight increase in PM₁₀ from condensation-mode aerosols is observed, but with much lower concentrations and a slightly different temporal pattern at the end of the episode.

The surface dynamics explained by RASPBERRY are supported by the vertical profiles from ALCs. Figure b reveals thick aerosol layers with high concentrations. Clouds prevented measurements above 2–2.5 km a.s.l, while also contributing to reduced mixing and increased concentrations at the surface. The vertical profiles of PM retrievals clearly show three distinct transport phases, reflecting the peaks in surface concentrations. Additionally, the depolarisation profiles (Fig. c) help identify the particle types throughout the episode: at the beginning and end of the episode, particles reaching the surface are irregularly shaped (medium to high depolarisation ratios), which is characteristic of mineral dust. In the middle of the episode, the particles are spherical (indicated by very low depolarisation ratios), which is typical of secondary aerosols. This episode highlights the value of physical source apportionment and active remote sensing techniques, particularly depolarisation-capable ALCs, in disentangling complex aerosol dynamics during mixed transport events.

The case of a summertime advection of secondary-rich aerosols is presented in Fig. S37. This coincides again with eastern surface winds, exceeding 8 m s⁻¹. The RASPBERRY retrievals highlight the reversed roles of condensation (higher concentrations) and droplet mode (lower concentrations) factors compared to winter, while vertical profiles emphasise the importance of the nighttime residual layer in contributing to surface concentrations .

In summer 2023, and to a lesser extent in 2024, record-breaking wildfires affected Canada , and on several occasions, their smoke was transported across the North Atlantic, reaching Europe . This phenomenon was accurately predicted by models e.g., and monitored from space by satellite radiometers . For example, Fig. a and b present aerosol optical depth (AOD) retrieved from the Moderate-Resolution Imaging Spectroradiometer (MODIS) onboard the Terra and Aqua satellites using the MAIAC algorithm, and carbon monoxide (CO) concentrations at 500 hPa from daytime AIRS/Aqua measurements on 27 June and 22 July 2023. These data clearly reveal high concentrations of both atmospheric constituents over Europe, and notably over the Aosta Valley (marked as a star in the plots).

Sudden increases in PM₁₀ concentrations were recorded at the surface on the same days. RASPBERRY attributed these increases to droplet-mode aerosols, which are rarely observed in summer (Fig. a). Conversely, the contribution by biomass burning-related aerosols and condensation mode particles remained negligible throughout the episodes (Fig. c and d). The attribution to the droplet mode can be explained by ageing processes during transport . It is important to note that condensation-mode aerosols were also present in the same period but showed distinct temporal patterns modulated by mesoscale circulation. For example, on 27 June (not shown) and in the morning of 28 June 2023 (Fig. S38), winds at all altitudes over Aosta originated from the north-western quadrant. In the afternoon of 28 June 2023, the wind direction shifted to the south-eastern quadrant, with back-trajectories clearly crossing the Po Basin. This change in air masses explains the anti-correlation between droplet- and condensation-mode behaviours, likely indicating the replacement of aged smoke particles with secondary (sulfate-rich) aerosols. A similar shift in circulation occurred on 22 July 2023, after the second event.

Although confirmation of the above interpretation based on chemical properties is not possible, as no chemical analysis was conducted in 2023, the dynamics are clearly supported by ALC profiles (Fig. e–h) and sun photometer retrievals (discussed later). The ALC measurements show that the surface PM₁₀ increases coincide with the rapid entrainment of elevated aerosol layers (> 4000 m a.s.l.) into the mixed layer near the surface. Not only is the timing of the droplet-mode concentration increases highly consistent, but the RASPBERRY-derived PM concentrations at the surface and those retrieved from the ALC (using the ALICENET algorithm) are of comparable magnitude (30–40 µg m⁻³ for the first event and approximately 20 µg m⁻³ for the second event). The particle depolarisation ratio measured by the CL61 ALC within the intruding layer is close to zero, further supporting the hypothesis of aged, nearly-spherical aerosols . It should be noted that aerosol depolarisation measurements obtained with the CL61 generally exhibit low signal-to-noise ratios, particularly for elevated and optically thin layers . Consequently, the highest smoke layers identified in the backscatter profiles (Fig. e–f) cannot be robustly characterised in terms of their depolarisation properties.

As for the measurements from the sun photometer, the AOD at 500 nm shows an increase during both events, with maximum values close to 0.8 on 27–28 June (first episode) and between 0.6 and 0.4 on 22–23 July (second episode). The extinction Ångström exponent (in the 400–1020 nm range) remains around 1.5 in both cases, never reaching the high values typical of fresh smoke from close sources. Single scattering albedo retrievals, consistently above 0.9 even at the shortest wavelengths, indicate only weak light absorption by the particles . The size distribution presents a maximum in the accumulation mode between 0.2 and 0.3 µm radius (i.e., 0.4–0.6 µm diameter), thus validating – on average along the atmospheric column – the particle size detected at the surface by the Palas Fidas and explaining the concentration increase in the droplet mode observed by the algorithm. These sizes are larger compared to the condensation mode usually present in summer (0.2–0.3 µm diameter), but are very consistent with published studies on aged smoke and notably on the same transport episode over Europe . In both episodes, neither the sun photometer nor the ALC provides evidence of other long-range transport phenomena such as desert dust.

A similar event occurred in August 2024 (Sect. S21 and Fig. S39) during an aerosol sampling campaign at high-altitude mountain stations in the Aosta Valley. Chemical analyses revealed a notable increase in organic compounds but negligible variation in secondary inorganics, and will be the subject of a future publication .

Interestingly, in all cases, droplet-mode PM₁₀ concentrations do not immediately return to baseline levels in the days following the peak, in contrast to their rapid initial increase. This persistence suggests that, once smoke has entered the mixed aerosol layer, it recirculates for several days before fully settling at the surface and being removed from the atmosphere, potentially posing implications for human health.