Vertical profiles of air constituents are particularly useful in identifying transport of pollutants of non-local origin. However, the profiling capability of the Italian regional environment protection agencies is still scarce. Over the Po basin, continuous monitoring of the atmospheric composition along the vertical profile is lacking and information at different altitudes is mostly available for short periods and during specific dedicated field campaigns e.g..

Light detection and ranging (lidar) instruments permit us to resolve the vertical distribution of particles. The recent technological and data-processing advances , and commercialisation, of simple lidar systems with operational capabilities allow to use this kind of system in monitoring (24/7) mode and in wide networks. In the present study, we employ two commercially available ALCs (CHM15k-Nimbus, manufactured by Lufft GmbH, and formerly by Jenoptik ESW), which have been operating since 2015 at the Aosta–Saint-Christophe observatory and in Milan. Both ALCs are part of the Italian ALICEnet (http://www.alice-net.eu/, last access: 28 February 2019) and the European E-PROFILE (https://ceilometer.e-profile.eu/profileview, last access: 28 February 2019) networks. They allow for continuous vertical profiling of the radiation emitted by a single-wavelength (1064 nm) pulsed laser (Nd:YAG; 6.5–7 kHz; 8 µJpulse-1) and backscattered by the atmosphere. At the operating wavelength, the backscatter is mainly dominated by aerosols and clouds in the atmosphere, whereas interference by water vapour has been estimated to be negligible . The systems enable a typical temporal resolution of 15 s (integration time) and a vertical resolution of 15 m, up to 15 km above the ground. The main limitations of the instruments are (1) the need for corrections in the lowermost levels and a (2) blind view above thick clouds. (1) In the lowermost levels, the field of view (0.45 mrad) of the receiver is only partially overlapped with the laser beam (90 % overlap is achieved at about 700 m); therefore, an overlapping correction function is needed to correct the signal. This was provided by the manufacturer. (2) Thick clouds cause saturation in the detector signal (an avalanche photodiode operated in photocounting mode), followed by complete signal extinction. Thus, the attenuated backscatter above the cloud ceiling is not considered (nor plotted) in this study. The ALC firmwares used so far (versions 0.730–0.743 for Aosta–Saint-Christophe and 0.730 for Milan) provide the background-, overlap-, and range-corrected attenuated backscatter (RCS) in terms of instrumental raw counts, i.e. RCS(z,t)=(P(z,t)-B(t))z2O(z), where P(z,t) is the signal intensity (raw counts) backscattered from a specific distance (z) and measured at ground, B(t) the time-varying background baseline, and O(z) the overlap function. To express the backscatter coefficient in SI units and make the results comparable with other similar instruments, a calibration factor (CL) must be assessed, so that RCS(z,t)CL=βatt(z,t)=βT(z,t)e-2∫zmin⁡zαT(z′,t)dz′, where βatt is the attenuated backscatter coefficient, βT is the total (particles and molecules) backscatter coefficient, and αT is the total extinction coefficient. CL is determined during clear-sky time windows of at least 3 h at night, i.e. when the background radiation is low, using the method (Rayleigh technique) described hereafter. First, the backscatter and extinction profiles are calculated with the Klett–Fernald backward algorithm ; then CL is determined by inverting Eq. (). Once a series of calibration factors has been estimated, the total (αT, βT) and particle (αp, βp) extinction and backscatter coefficients are computed for all times and sky conditions using a forward Klett method as described by .

Usually, the above-mentioned solving techniques are based on an a priori or independent estimate of the lidar ratio (LR, i.e. the ratio αp/βp) as a further constraint. In our case, LR is not fixed a priori but rather obtained using specific functional relationships linking αp to βp. demonstrated that this approach, previously proposed and tested on the signal inversion of research-type elastic lidars e.g., provides better retrievals of αp and βp also from ALCs than using an a priori fixed LR. More specifically, an iterative data inversion scheme is adopted: at the first iteration, LR is set to an initial value of 38 sr (average value from the functional relationships) and a first retrieval of the backscatter coefficient βp is calculated; starting from the second iteration, the calculated backscatter coefficient and the functional relationships are used to determine an altitude-dependent lidar ratio. The loop continues until convergence of the column-integrated backscatter is reached. The good agreement between the ALC-derived and the sun-photometer-measured aerosol optical depth (AOD, i.e. the integral over altitude of the extinction coefficient) is employed as a validation of the quality of the inversion results (Sect.  and ) using these functional relationships, at least in daytime conditions (see also , and Figs. S2d and S9e in the Supplement).

For ease of comparison with pristine (aerosol-free) conditions, and with most lidar-based studies, ALC measurements are provided in this study in terms of scattering ratio SR, e.g., i.e. SR=βTβm=βp+βmβm, where βm is the molecular backscatter coefficient. In the case of pure molecular scattering (no aerosol in the atmosphere), SR=1, while SR increases with increasing aerosol load. Finally, the high-resolution data from the ALC are downscaled to 75 m averages over the vertical and 5 min averages over time to increase the signal-to-noise ratio. A first example of the output from the Aosta–Saint-Christophe ALC, in terms of scattering ratio, can be found in Fig. a. This image refers to a typical sequence of days (August 2015), characterised by a recurrent increase in the particle backscatter during the afternoon, up to an altitude of more than 2000–3000 m a.s.l. (the altitude of the surface being 560 m a.s.l. in Aosta–Saint-Christophe). As we will demonstrate here, this afternoon increase is due to the coupled effect of transport of polluted air masses from the Po basin and aerosol hygroscopic growth (see Sect. ). Several analogous episodes were recorded in the ALC record since its installation in Aosta–Saint-Christophe. The observation of this recurrent phenomenon was, in fact, the driving motivation for the present research.

In the study, we also convert the ALC-derived backscatter into aerosol volume following , thus allowing a direct comparison to more standard air quality metrics (e.g. PM10; this is carried out using a particle density ρ=1.3 gcm-3 independently estimated for the present study by an OPC co-located with the ALC). The expected uncertainties in the retrieval of the aerosol backscatter and extinction coefficients and of the aerosol volume range between 30 % and 40 % .

A POM-02 sun–sky radiometer has operated at the Aosta–Saint-Christophe observatory since 2012. The radiometer is part of the European ESR-SKYNET network (http://www.euroskyrad.net/, last access: 28 February 2019). The irradiances collected by the POM-02 at 11 wavelengths (315–2200 nm) are inverted to retrieve the aerosol optical properties using both the direct sun (SUNRAD.pack algorithm to provide the AOD every 1 min, ) and the almucantar geometries (SKYRAD.pack software version 4.2 to retrieve a complete set of optical and microphysical columnar properties every 10 min, ). The instrument is calibrated in situ with the improved Langley technique, described by in more detail, and was successfully compared to other reference instruments during a recent international campaign . The AOD (τ) from the POM-02 is interpolated to the ALC wavelength (1064 nm) using the relationship, i.e. τ=bλ-a, where λ is the wavelength expressed in micrometres and a and b are the Ångström parameters from the regression. Finally, the Cloud Screening of Sky Radiometer data (CSSR) algorithm by , making use of the short-wave irradiance measurements by a co-located pyranometer, is applied to the POM-02 series to minimise the residual interference by clouds and to ensure the maximum measurement quality.

In this study, we also used satellite data to explore if and how the “local” phenomenon observed in Aosta is detectable over a regional scale. To this purpose, we used AOD data from the Moderate Resolution Imaging Spectroradiometer (MODIS) instrument. The MODIS instrument flies on board the two NASA platforms Terra and Aqua, following a sun-synchronous orbit with overpass times between 10:00 and 13:00 and 13:00 and 16:00 (local time), respectively. Since the MODIS instrument planning phase, specific retrievals have been set up to provide the AOD over ocean and land globally on a daily basis at 10 km resolution . Constant improvements to the AOD inversion algorithms resulted in a 3 km resolved standard AOD product . While such spatial resolutions have been extensively exploited for many regional-scale, aerosol-related studies, these are yet not sufficient for applications requiring more spatial detail, as in space-based evaluations of air quality within urban areas e.g. or in conditions of high AOD spatial variability as over mountain regions e.g.. For our purpose, we therefore used high-resolution (1 km) AOD data obtained inverting MODIS data with the recently developed algorithm MAIAC (Multi-Angle Implementation of Atmospheric Correction). Full details of this algorithm are thoroughly described in .

A Fidas^® 200s OPC operates at the ARPA observatory in Aosta–Saint-Christophe. The spectrometer is based on the analysis of scattered light at 90∘ originating from a polychromatic light source (LED). These conditions ensure an accurate calibration curve without ambiguities within the Mie range and allow us to retrieve high-resolution spectra (size measurements between 0.18 and 18 µm, 32 channels decade-1). Due to the peculiar T-aperture optics of the spectrometer and the simultaneous measurement of signal duration, border zone errors are eliminated. Once the particle size distribution is measured, the instrument algorithm is able to derive the mass concentration for several cutoff diameters (including PM10, i.e. ambient particulate with a diameter of 10 µm or less). Though not a direct mass measurement, the PM concentration derived by the instrument obtained the certificate of equivalence to the gravimetric method by TÜV Rheinland Energy GmbH on the basis of a laboratory test and a field test. Moreover, to prevent any site-specific bias, an additional PM10 comparison with the gravimetric technique was organised at Aosta–Saint-Christophe and provided satisfactory results (29 days; slope 1.08±0.04; intercept -3.8±1.4 µgm-3; R2=0.96).

Daily averages of PM concentration are recorded by four Opsis SM200 particulate monitor instruments, two in Aosta–downtown (PM10 and PM2.5 inlets, with sampling fluxes of 1 and 2.3 m3h-1, respectively), one installed inside a mobile laboratory, which was parked in Antey (PM10, 1 m3h-1), and one in Donnas (PM10, 1 m3h-1). Moreover, two tapered element oscillating microbalance (TEOM) 1400a monitors are used for continuous measurements of PM10 hourly concentrations at the stations of Aosta–downtown and La Thuile. These instruments do not compensate for mass loss of semi-volatile compounds and could be insensitive to specific compounds, such as ammonium nitrate e.g., which leads to underestimations, especially in the cold season, compared to the SM200. Conversely, overestimations by the TEOM compared to daily averages from the SM200 reference instrument are found in summer and are not fully understood at present. Therefore, TEOM monitors are only employed here for qualitative estimates of short-term variations in the aerosol burden while daily-averaged concentrations will only be taken from the SM200 instruments.

Sampling in Aosta–downtown is complemented with chemical speciation analyses. We employed a Dionex ion chromatography system (AQUION/ICS-1000 modules) for water-soluble anion–cation chemical analyses on daily PM10 samples collected on PTFE-coated glass fiber filters by the Opsis SM200. The experimental setup is based on the CEN/TR 16269:2011 guideline and enables the determination of mass concentrations of the following water-soluble ionic compounds: Cl-, NO3-, SO42-, Na+, NH4+, K+, Mg2+, and Ca2+. Samples collected on quartz fibre filters by a co-located Micro-PNS automatic low-volume sampling system (10 µm cutoff diameter, 2.3 m3h-1) are analysed alternatively for elemental/organic carbon (EC/OC, 4/10 days) and for metals (6/10 days; not used in the present study, but discussed by ). The carbonaceous aerosol mass is determined with a Sunset Laboratory Inc. instrument on portions of 1 cm2 punches using a thermal–optical transmission (TOT) method with transmission correction for the split point and following the EUSAAR-2 protocol , according to the EN 16909:2017.

Standard gas-phase pollutants subject to European regulations are routinely monitored at Aosta–downtown, La Thuile, and Donnas in the frame of the activities of the air quality network. Meteorological parameters, such as temperature, pressure, RH, and surface wind velocity are collected at the stations of Aosta–Saint-Christophe, Saint-Denis, and Donnas. Moreover, 10 temperature and RH sensors (Hobo H8 Pro) are installed along the north-facing mountain slope south of Aosta, at elevations ranging from 550 to 1200 m a.s.l. This set of measurements, representing a vertical profile of surface temperature and RH, gives useful information about the thermal inversions in the main valley. For example, pseudo-equivalent potential temperatures e.g. at different altitudes can be easily calculated from this dataset, thus providing a rough indication of the vertical extent of the mixed layer (Sect. ).

COSMO is a non-hydrostatic, fully compressible atmospheric prediction model working on the meso-β and meso-γ scales. A detailed description of the model can be found elsewhere (e.g. ). The COSMO data are operationally disseminated by the meteorological operative centre – air force meteorological service (COMET) in two different configurations: a lower-resolution (7 km horizontal grid and 45-level vertical grid, 72 h integration) version (COSMO-ME), covering central and southern Europe, and a nudged, higher-resolution version (2.8 km, 65 vertical levels, 2 runs day-1), called COSMO-I2 (or COSMO-IT), covering Italy (Fig. ). Owing to the complex topography of the Aosta Valley and the consequent need to resolve the atmospheric circulation at very small spatial scales, the COSMO-I2 variant is employed in this work.

As an example of the good agreement between COSMO and surface measurements, the average daily cycle of the wind speed and direction from both data sources is exhibited in Fig. S1 in the Supplement. The figure clearly shows the regular development of the plain–mountain winds in the afternoon. The influence of the east–west directrix of the main valley along which the wind is channelled is well represented in both measurements and simulations.

FARM v4.7, is a three-dimensional Eulerian model for simulating the transport, chemical conversion, and deposition of atmospheric pollutants. The FARM source code has been inherited from the Sulfur Transport and dEposition Model (STEM), extensively tested and used since the 1980s. FARM can be easily interfaced to most available diagnostic or prognostic NWP models. A turbulence and deposition pre-processor (SURFace-atmosphere interface PROcessor, SURFPRO) computes the 3-D fields of turbulence scaling parameters, eddy diffusivities, and deposition velocities for each species based on an input gridded land-use field and the results of the NWP model . Pollutant emissions from both area and point sources can be simulated by FARM including plume rise calculations. Transformation of chemical species by gas-phase chemistry (more than 200 reactions using the SAPRC-99 chemical scheme as in ), dry removal of pollutants depending on local meteorology and land-use, and wet removal are considered. The AERO3_NEW module, coupled with the gas-phase chemical model and treating primary and secondary particle dynamics and their interactions with gas-phase species, is implemented for the calculation of the aerosol concentration fields, thus accounting for nucleation, condensational growth, and coagulation . The aerosol size distribution is parametrised using three modes simulated independently: the Aitken mode (D<0.1 µm), the accumulation mode (0.1 µm <D<2.5 µm), and the coarse mode (D>2.5 µm). PM2.5 is defined as the sum of Aitken and accumulation modes, while PM10 is given by the sum of the three modes. Chemical speciation is performed in the pre-processing phase by the emission manager (EMMA) based on the profiles from the U.S. EPA model SPECIATE (v3.2, 2002; see https://www.epa.gov/air-emissions-modeling/speciate-version-45-through-40 for more recent versions, last access: 28 February 2019). To simulate hygroscopic growth by aerosols in high-RH conditions, water uptake by aerosol particles is taken into account based on the ISORROPIA model and added to the PM10 dry mass concentration from FARM. The resulting output species is called PM10w in FARM version 4.7.

The FARM output concentrations are 4-D fields at 1 km spatial resolution along the horizontal dimensions, 16 different vertical levels (from the surface to 9290 m, corresponding to equally spaced pressure levels), and 1 h temporal resolution. The PM10w concentration profiles from FARM are extracted at the grid cell corresponding to Aosta–Saint-Christophe for comparison with the profiles measured by the ALC. Indeed, since FARM is not able to calculate the aerosol optical properties needed to simulate the backscatter coefficient measured by the ALC, the comparison between the profiles measured by the ALC and estimated by the chemical transport model (CTM) is performed here in terms of mass concentration (by converting the ALC data into PM10; see Sect. ).

Supplying a detailed and precise emission inventory to the CTM is crucial to accurately assess the magnitude of the pollutant loads and their variability in both time and space. Additional information regarding the regional emission inventory and the boundary conditions is provided in the Supplement (Sects. S2–S3). The geographic coverage of the regional and the national emission inventories is shown in Fig. . Local sources and boundary conditions can be switched on or off for sensitivity analyses.

The publicly available LAGRANTO Lagrangian analysis tool, version 2.0 , is used to numerically integrate the high-resolution 3-D wind fields from COSMO and to determine the origin of the air masses sampled by the ALC over Aosta–Saint-Christophe. The software also enables to trace 3-D and 2-D meteorological fields along each trajectory. In particular, the algorithm was set up to start eight trajectories in a circle of 1 km around the observing site and at seven different altitudes from the ground to 4000 m a.s.l., for a total of 56 trajectories for every run. From a 1-year (2016) analysis of the trajectories arriving to the Aosta Valley, it is found that a backward run time of 48 h is sufficient, on average, to cover most of the domain of the meteorological model. Therefore, we limit the computation to this duration.

A thick aerosol layer is detected by the ALC over the Aosta–Saint-Christophe observatory from the afternoon of 26 August (Fig. a). Very clear conditions are visible in the low troposphere on 25 August, while the high-altitude layer in the morning of the same day, which is not considered here, is due to smoke transport from North America. The appearance of the PBL layer is clearly noticeable on 26 August as an increase in the backscatter coefficient up to an altitude of 3 km a.s.l., with scattering ratios SR ≃4 at midday (light blue area in the figure) almost doubling (SR >8, orange yellow) in a few hours. This layer persists during the night, when the SR reaches values above 30. On 27 August, the ALC backscatter is then observed to decrease in the central part of the day and to increase again in the afternoon. This behaviour stays very regular for almost a week, with the aerosol-rich layer extending from the ground up to 3–3.5 km. As further discussed in the next paragraphs, we anticipate here that the main factor driving this cycle is likely an enhanced hygroscopic growth of aerosol advected from the Po Valley from the afternoon to the early morning, with this effect also leading to the formation of low clouds within the aerosol layer at night (screened out as white areas in the figure). In fact, the transition from aerosol to the cloud phase is very sharp, as also noticeable from the sudden increase of more than 40 Wm-2 in the downward infrared irradiance monitored at the same site (Fig. S2c).

Simultaneous ALC measurements in the city of Milan (see relative position in Fig. ), which can be considered representative of the overall dynamics occurring within the Po basin, are shown in Fig. . An interesting feature here is that the modulation of the scattering ratio looks almost reversed compared to the Aosta–Saint-Christophe site, with a maximum SR at the surface at midday and minimum values during the night and the morning. While in the uppermost levels (>3000 m a.s.l.) the synoptic circulation is blowing undisturbed from the west, the wind velocity at 500 m a.s.l. keeps alternating, likely driven by the breeze regime (the surface wind is affected by urban effects and does not show appreciable variations).

The observed reversal behaviour in Milan and Aosta already suggests that air mass movements are driving the clean-up of the lowermost levels in the Po plain and the transport of the aerosol plumes elsewhere. To substantiate this hypothesis, a careful analysis of the meteorological fields (observed and modelled) was performed. In particular, we verified that this selected sequence of days presents a typical pattern of plain-to-mountain wind systems during the afternoon of each day in Aosta–Saint-Christophe. Surface-level eastern wind speeds as high as 8 ms-1 are measured daily in the afternoon till sunset and are shown as bold arrows in the lowermost levels of Fig. a. Conversely, calm wind is detected during the night, i.e. when the aerosol layer thickens. Since no instrument is available at the measuring site to determine the vertical profile of the wind velocity, the simulations from the COSMO model are used to assess the wind field at several altitudes (thin arrows in Fig. a). It reproduces the thermal wind circulation in the lowest atmospheric layers well during the afternoon and slightly overestimates the mountain-to-plain drainage winds at night and early morning (this issue is discussed in Sect. ). The thermally driven wind pattern forecasted by COSMO extends up to an altitude of 3000 m, i.e. approximately the maximum height of the aerosol layer observed by the ALC (reasons for possible discrepancies of this simulated and measured altitude are discussed in Sect. ). Note that wind direction is incompatible with Aosta being the potential source of the observed aerosol layer, as the city is located west of the observatory. At higher elevations, the wind field is clearly decoupled from that in the PBL and follows the large-scale circulation.

Complementary information is provided by the analysis of the 48 h back trajectories calculated by LAGRANTO using COSMO fields (Sect. ) ending over the Aosta Valley in the period addressed (Figs.  and S3). For ease of clarity, the LAGRANTO output is shown in separate panels depending on the arrival altitude of each trajectory (<2000 and >2500 m a.s.l.). These results show that before the episode (25 August and morning of 26 August, Fig. S3a and d) trajectories are driven by large-scale flows from the west and are thus parallel at all altitudes. Therefore, air masses reached Aosta after crossing the Alps, notably the Mont Blanc chain, hence transporting clear and unpolluted air from the free troposphere to the PBL. Then, in the afternoon of 26 August, back trajectories in the PBL change their provenance owing to the development of the thermal circulation tapping into air masses of very different origin (Fig. a), while higher-altitude trajectories mostly continue to follow the synoptic circulation (Fig. b). The lowermost trajectories cover a notable distance and cross some major conurbations of the Po basin, i.e. Milan and Turin, at altitudes lower than a few hundred metres above sea level, and thus well within the polluted PBL. This sudden reversal of the trajectories occurs simultaneously with the appearance of the elevated aerosol layers in the Aosta ALC image (Fig. a). These meteorological conditions persist for the rest of the day and in the following days. The analysis of the corresponding back trajectories confirms that transport of polluted air masses from the Po basin also occurs in the afternoons of the other days of this episode, until the flux changes again to a northwestern configuration (Fig. S3c and f).

To complete the picture, it is worth mentioning that the COSMO model also predicts an increase in the RH (Fig. b) from evening to early morning, almost simultaneous with the SR enhancement observed by the ALC. In this time frame, RH exceeds typical summertime deliquescence values reported for the Po basin in previous studies e.g. DRH =67 %, and reaches up to 98 % at the ground (Fig. S2b). This suggests hygroscopic growth on aerosols and a consequent increase in the ALC βp. For example, at a measurement site representative of Po Valley conditions, found a median increase in the aerosol backscatter coefficient of 70 % for RH =90 % compared to the dry case. During the day, RH decreases below typical crystallisation values in summer e.g. CRH =62 %,. As RH is clearly modulated by the temperature daily cycle, the measured specific humidity (SH) is also plotted on the same figure (Fig. S2b) as an additional variable independent of temperature, to identify potential advection of different air masses to the observation site. Indeed, a SH increase occurs on 26 August (starting from minimum values of ∼8 gkg-1 in the morning to about 11 gkg-1 in the evening) as soon as the wind starts blowing and high values (>13 gkg-1) endure for the rest of the week. This likely indicates that the dry air, typical of the more mixed mountain PBL , is replaced by more stagnating, and humidified, air masses characteristic of hot summer days in the Po Valley . This scenario is compatible with recent findings by , who performed water vapour measurements with the POM-02 at Aosta–Saint-Christophe and found that moist air masses mainly come from the east. A discussion about the constancy of the measured SH during each day compared to the more variable values forecasted by COSMO is provided in Sect. . We anticipate here that this behaviour confirms that COSMO overestimates the nighttime drainage winds (characterised by lower SH) and contributes to the observed discrepancies between the PM concentrations from FARM and the ALC.

We also show in Fig. c the altitude-resolved aerosol mass derived from the ALC backscatter coefficient (as described in Sect. ). The maximum concentration within the aerosol layer is >80 µgm-3. The corresponding PM10w profile from FARM, partitioned between the non-local (coloured background) and local (contour line) pollution, is shown in Fig. d. FARM qualitatively reproduces the recurrent increase in the aerosol concentration at the end of each day and mainly ascribes it to particles transported by the thermal winds from the model-box boundaries. As an example, Fig.  provides a 3-D snapshot of the model simulation results, clearly showing the entrance of the aerosol-rich air mass from the Po basin to the Aosta Valley. The picture refers to 28 August 2015 at 15:00 UTC – the whole sequence of 26–31 August 2015 is available as a video file in the Supplement (10.5446/38391). Still, there are two important differences between the FARM model simulations and the ALC observations in terms of (1) absolute PM10w concentrations and (2) timing of the phenomenon. In fact,

PM10w values from FARM are much lower than the ones retrieved from the ALC (about -40 % outside the thick aerosol layer identified by the ALC at night and even -80 % inside the layer);

Possible reasons, such as hygroscopicity effects and modelling deficiencies, explaining the above-mentioned issues are further discussed in Sect. .

To evaluate the impacts on surface air quality parameters during the episode, hourly PM10 concentrations at the surface as measured in Aosta–downtown and simulated by FARM in Aosta–Saint-Christophe are presented in Fig. e (PM10 monitoring at La Thuile was not yet operational at that time). Apart from two spikes (80 µgm-3) on 25 and 26 August (presumably of local origin), the concentrations measured in Aosta–downtown by the TEOM show a slight increase after the arrival of the layer, but without sudden jumps. Also, PM concentrations are generally higher during daytime compared to the night, according to the expected cycle of the summertime local sources (e.g. traffic, resuspension). These features, however, can be connected to the fact that mass loss occurs in TEOM due to secondary aerosol volatility, as better discussed in the companion paper by comparing the daily PM10 cycle from this instrument and the Fidas OPC in Aosta–Saint-Christophe. Moreover, this volatility could be different between nighttime and daytime, which would also contribute to the observed daily behaviour. In addition, FARM estimates at the surface are again lower than measurements (-60 %, on average). Daily PM10 concentrations observed with Opsis SM200 instruments during the case study in Aosta–downtown and Donnas are shown in Fig. , which includes the whole episode (correlation index with TEOM measurements ρ=0.84). The shaded area corresponds to those dates affected by the thick layers as revealed by the ALC. An increase in daily concentrations (up to maximum values of 10 µgm-3 for PM2.5 and 16–22 µgm-3 for PM10) can be clearly noticed at both sites, leading to concentrations slightly higher than average for the same period (7 µgm-3 for PM2.5 in Aosta–downtown and 12 µgm-3 for PM10 at both sites, considering the 2015–2017 series). The statistical significance of such PM increases during these transport episodes, compared to the natural variability in the aerosol load in non-advection conditions, is assessed in the companion paper using the full dataset. The daily averages of the simulated aerosol concentrations at the surface are superimposed on the same figure (dashed lines). While the model qualitatively reproduces the average load of PM2.5 and its variations in Aosta–downtown, it underestimates PM10 at both stations as already noticed.

Since sun photometric measurements only can be performed in daylight, results are often unavailable at those times when the ALC shows the greatest backscatter signal, i.e. in the evening and at night. However, data collected by the POM-02 radiometer can still be effective to monitor the first (late afternoon) and last (early morning) dynamics of the aerosol layer as seen by the ALC (Fig. a), and particularly tell us if this signal is detectable in the sun-photometer-derived, column-integrated aerosol load. AOD obtained from the sun photometer (Fig. S2d) varies from 0.02 (25 and 26 August before appearance of the layer) to 0.07 (29 August, morning) at 1064 nm (approximately 0.05 to 0.2 at 500 nm) and closely follows the AOD obtained by vertically integrating the extinction coefficient from the ALC over the atmospheric column. The two independent AOD retrievals present a mean bias of -0.007 and standard deviation of the differences of 0.006, both lower than the declared uncertainty of the POM sun photometer itself (about 0.01) . The good closure with the AOD from the photometer demonstrates the reliability of the functional relationships derived by and employed in our ALC inversion algorithm, at least during the daytime.

Further retrieval products from SUNRAD.pack and SKYRAD.pack (displayed in Fig. S2e) show the Ångström exponent to increase from 1.2 to 1.7 on 26 August from 08:00 to 17:00 UTC, suggesting the advection of smaller particles in the atmosphere, and to remain almost constant (about 1.6, a typical value for the Po Valley, as already described by , and ) in the following days. These values should be compared to the lower Ångström exponents typically measured in the Aosta Valley, i.e. ∼1.1 on average . Likewise, the single-scattering albedo (SSA) at 500 nm increases (from 0.7 to 0.95) on 26 August, which is compatible with the arrival of more scattering (likely secondary aerosol, as described in Sect.  and ) and/or more aged aerosol, such as that from the Po Valley . The sun-photometer-derived total-column aerosol volume distribution (Fig. S2f) peaks in the accumulation mode (about 0.3 µm). A slight decrease in the peak diameter in the morning (from about 0.4 to 0.2 µm) can be noticed on some days (e.g. 27–30 August) and might be ascribed to the dehydration of the particles as temperature increases and RH decreases. The same behaviour can be observed better in the third case study (Sect.  and Fig. S9g).

In order to provide a first evaluation of whether the phenomenon observed and described in detail for the Aosta area could have a more general validity in the Alpine region, we used AOD data retrieved from space over northern Italy. In particular, we exploited the high-resolution capabilities of the MODIS–MAIAC AOD product (Sect. ) and the availability of two MODIS overpasses during the day (Terra and Aqua platforms), to detect signs of the described effects at the regional scale. Figure a shows the average difference between the AOD retrieved each day from MODIS Aqua (overpass time between 12:00 and 13:00 UTC) and that from MODIS Terra (10:00–11:0 UTC). Despite the short time lag between the Terra (AM) and Aqua (PM) satellite overpasses, this figure shows that the data are sufficient to start detecting an overall reduction of the AOD in the Po basin (blue area) and a reverse increase in the mountain areas (Alps and Apennines) surrounding it. The general picture suggests a sort of aerosol drainage from the Po Valley (negative AOD difference, blue) to the Alps (positive AOD difference, red), although some aerosol dehydration from the morning to the afternoon could also partially contribute to the observed morning and afternoon differences. This provides an observation-based confirmation of the hypothesis of aerosol transport, in agreement with our previous results from FARM (e.g. Fig.  and the relative video file), and with wind simulations from COSMO over the same area (averaged over the same hours between Terra and Aqua overpasses, Fig. b). Valley–mountain (and sea–land) breezes are clearly reproduced, as expected on days with weak synoptic flows and strong heating by the sun.

The profiles from the ALC in Aosta–Saint-Christophe for this winter case are depicted in Fig. a and show the sudden appearance of a thick aerosol layer in the afternoon of 26 January. Unlike the previous case, the ALC measurements do not reveal distinct features for each day of the sequence, but rather a continuous and persisting layer during the whole episode. The SR reaches values above 30 in the night between 26 and 27 at altitude, and, closer to the surface, between the evening of 27 and the morning of 29 January. The layer extends up to 2000 m a.s.l., a clear signature of the non-local origin of the air mass. Some clouds are visible above and within the aerosol layer. The episode ends on 29 January as quickly as it began, with clearer air taking the place of the polluted air mass starting from above and subsequently eroding the layer down to the surface.

Simultaneous ALC profiles over Milan are depicted in Fig. . As opposed to the Aosta Valley, the aerosol layer does not vanish on 29 January, but remains for a few days longer, although the winds at altitude change their provenance from the west on that day. Clouds only form from 27 January, presumably allowing solar radiation to trigger a weak breeze tide in the lowest 2000 m on that day, whilst strong stability favours calm wind in the following days.

The wind field over Aosta–Saint-Christophe, depicted in Fig. a, presents a very different pattern compared to the first case addressed (Fig. a). Firstly, calm wind is measured for the whole period at the bottom of the valley. This is due to a shallow temperature inversion in the lower atmospheric layers in the main valley. Conversely, at the Saint-Denis station, located above the inversion layer, and at the Donnas station, where the temperature inversion is weaker, the wind pattern is more representative of the wider circulation: for example, the average wind speed in Saint-Denis is about 4 ms-1 on 26 January in the afternoon (Fig. S4e) and the wind clearly turns from west (morning) to east (afternoon), simultaneously with the appearance of the layer. The same wind change is detected in Donnas on the same day (Fig. S4f), with easterly wind speeds >1 ms-1 for several hours in the afternoon. As a further difference with the first case, the forecasted wind at 1000–2000 m a.s.l. does not show any change in direction typical of the thermal winds. For example, at 2000 m a.s.l. the circulation is continuous and vigorous (up to 6 ms-1), from the afternoon of 26 to the beginning of 29 January. Indeed, this winter case study interestingly shows that thermally driven winds are not the only mechanism, especially in winter, driving the advection of air masses from the Po Valley to the Alps. Rather, the synoptical circulation can push the air masses towards the Alpine valleys, as in this case. In fact, the flow clearly reveals its southern origin at elevations above the mountain crest (e.g. 3000 m a.s.l.), where the wind is not channelled within the main valley. At that altitude, the wind speed is even greater than 20 ms-1. Finally, on 29 January, the measurements in Saint-Denis (gradual increase in the speed of westerly wind) and in Donnas (even stronger wind, again from the west), and COSMO simulations (wind reversal at 1000–2000 m a.s.l.) correlate with the disappearance of the layer better than observations performed at the bottom of the valley (calm wind).

Back trajectories for 26 January are plotted in Fig. S6 and indicate transit over the Po basin starting from the morning (panels a, d), which seems to contradict the fact that the layer arrival over the Aosta Valley is detected by the ALC only since the afternoon. This can be explained by noting that the mean altitude of the trajectories crossing the Po basin during the morning exceeds 1500 m a.s.l. and is thus higher than the Po Valley aerosol layer observed by the ALC in Milan (Fig. ). The trajectory altitude tends to decrease in the afternoon, reaching the elevations of the polluted boundary layer (Fig. S6b) and thus leading to effective aerosol transport to the Aosta Valley. In fact, considering that each dot in Fig. S6b represents a 1 h step, we estimate a mean air mass residence time in the Po Valley PBL of 30–35 h before arriving over the observing site. Finally, trajectories turn westerly on 29 January, in agreement with the removal of the layer over Aosta (Fig. S6c and f).

Together with the appearance of the aerosol layer, an increase in the COSMO RH can be noticed (Fig. b). The latter remains higher, above typical wintertime deliquescence values e.g. DRH =54 %,, for the whole duration of the episode, and never drops below the crystallisation point (e.g. CRH =47 %), which can also be partly attributed to the presence of low clouds forecasted by the NWP model, as actually occurred. The advection is detected more clearly by the increase in SH measured at ground level (from less than 2 gkm-1 to a maximum of 4 gkg-1 on 28 January, Fig. S4b). For this episode, the arrival of a different air mass is additionally revealed by the temperature and humidity sensors along the mountain slope. Pseudo-equivalent potential temperatures at different altitudes are shown in Fig. S5. As clearly noticeable, the spread among the series recorded at 550 m a.s.l. and that at higher altitudes remarkably, and quickly, decreases on 26 January, especially during the night, suggesting that the strong (and very shallow) temperature inversion weakens and mixing of the upper aerosol layers down to the surface is favoured.

The mass concentration retrieved within the layer by the ALC (Fig. c) is quite variable (from 30 µgm-3 at the edge of the layer to more than 100 µgm-3 at the core) and reveals the heterogeneous distribution of the particulate inside the layer. FARM predicts a very different scenario, with three separate increases at the end of 26, 27, and 28 January of non-local origin (coloured background in Fig. d, much lower than the concentration retrieved by the ALC) and a clear diurnal cycle close to the surface of local origin (Fig. e). The diurnal cycle in the simulations is characterised by two peaks corresponding to the combined effect of traffic rush hours, residential heating, and variation in the mixing layer height. Hourly and sub-hourly PM10 surface concentration measurements at both Aosta–downtown and Aosta–Saint-Christophe, however, only exhibit one peak at midday. Taking into consideration these different daily evolution patterns and the sources included in the emission inventory, the most likely reasons for the differences between the model and the measurements at the surface appear to be an underestimation of the residential heating (actually switched on all day during these very cold days) and an overestimation of the traffic road contribution, together with an overestimation of the mixing layer height growth at midday by the NWP model. Anyway, Fig.  shows that the daily averages of PM concentrations measured at several sites of the region are higher on 27–28 January than on the neighbouring days. Specifically, the increase is similar (more than 40 µgm-3) for both PM2.5 and PM10 in Aosta–downtown, which results from the fact that the increment is mainly driven by particles with a diameter of less than 2.5 µm. The maximum PM10 concentration (117 µgm-3) was measured on 28 January in Donnas (Fig. h), which is the closest station to the Po basin. The spatial pattern of the observed increase, not fully captured by the model, is evident in Fig. S7 and represents a further indication of the Po Valley being the source of the polluted air masses. Moreover, Fig. c, d show this increase to be associated with enhancement in Aosta–downtown of the nitrate and ammonium components (see next paragraph), two key species of the Po Valley secondary aerosol, but minor contributors to PM10 in the Aosta Valley. Finally, while the daily PM concentrations from FARM are comparable, on average, to the measurements, the modulation of the PM concentration by the advection (peaks) is not captured by the model, whose output is rather constant. Most interestingly, data collected at remote and usually pristine sites also show a remarkable increase: at La Thuile (PM10 winter average 7 µgm-3), the hourly PM10 concentration (Fig. e) reaches nearly 40 µgm-3 (some hours later than the appearance of the aerosol layer in Aosta–Saint-Christophe) and correlates well with the increasing NO2 concentration (from about 2 µgm-3 before and after the event to 44 µgm-3 during the event on an hourly basis) measured by a co-located detector. Additionally, the mobile laboratory in Antey (winter average 20 µgm-3) measures increasing daily PM10 concentrations with a maximum of 69 µgm-3 on 27 January (Fig. g) and increasing NO2 concentrations from about 30 to 56 µgm-3.

For this selected sequence of days, the data collected by the OPC in Aosta–Saint-Christophe are additionally available. The instrument reveals a notable increase in the number concentration for particles smaller than 0.5 µm (Fig. S4c) in coincidence with the arrival of the aerosol layer. The total number concentration (Fig. S4d) gradually increases from a few hundred up to 3000 particles cm-3 and decreases again on 29 January (the average value in winter 2016–2017 in conditions of local pollution being 650 particles cm-3).

Some results of anion–cation analyses performed on daily samples collected at Aosta–downtown are also reported in Fig.  and presented in terms of relative concentrations (ratio between ion mass and PM10). As anticipated, the fractions of nitrate and ammonium drastically increase during the event, reaching values more than double (nitrate) or even 8 to 10 times as much (ammonium) compared to the concentrations in the days adjacent to the case study. Indeed, wintertime low temperature and high humidity in the Po Valley represent the best conditions leading to the formation of ammonium nitrate . In addition, this nitrate increase enhances the observation of a lowering of DRH that may influence the ALC backscatter. Sulfate also increases, but not as much as nitrate and ammonium since unfavourable conditions are met during winter . Only one sample was analysed for EC and OC during the event, and the OC/EC ratio increases only marginally, likely due to sample overloading. In general, variations in the aerosol composition are noticeable on 27–28 January and in line with transport from the Po basin (as more rigorously demonstrated using the positive matrix factorisation method in the companion paper). Indeed, a high presence of secondary aerosol, most notably nitrate compounds, in the Po Valley has been documented for a long time . These are probably enhanced by the particular atmospheric conditions during the examined period . All together, nitrate, ammonium, and sulfate can explain about 40 % of the PM10 mass during the episode (as a reference, this fraction represents 15 % of the PM10 mass for non-advection days of January–February 2017, on average, owing to missing sources of precursors in the Aosta Valley), while organic matter (OM, assuming a typical conversion factor of 1.6 between the measured concentration of OC and the unknown concentration of OM; as in and for urban sites) and EC account for the remaining 30 % and 5 % fractions, respectively (similar percentages are obtained for non-advection days in January–February 2017). Finally, the relative concentration of the other measured ions, allegedly of local origin (e.g. Na+ and Cl- from road salting, not shown), does not follow the same pattern as observed in Fig. . Figure c–e also reveal that FARM is not able to reproduce the experimental chemical speciation: nitrate is strongly underestimated, while ammonium and sulfate are strongly overestimated, and the simulations of the OC/EC ratio do not follow the experimental data. This behaviour is probably to be ascribed to the fact that the SPECIATE v3.2 chemical characterisation implemented in the emission manager is not suitable for the considered sources and/or that the sources, and therefore their chemical profiles, are not accurately identified.

Since the establishment of the thermally driven wind regime, starting from 25 May, a thick aerosol layer is regularly detected by the ALC in the afternoon (Fig. a). The layer persists during each night, when the scattering ratio increases up to a value of 20 and clouds systematically form within the layer. This aerosol layer extends from the ground to an altitude increasing from 2.5 km at the beginning of the case study to more than 3 km at the end of the episode.

The plain-to-mountain circulation, driving the phenomenon under investigation, is well captured by both measurements at the surface (Fig. a, bold arrows) and COSMO forecasts (thin arrows). Eastern winds with speeds >10 ms-1 are measured in the afternoon till sunset at the surface, while nights are characterised by calm wind. At higher elevations, the wind provenance turns from the north, at the start of the depicted sequence, to the south.

The back trajectories ending over the Aosta Valley during the third episode are plotted in Fig. S8. On 25 May, the large-scale circulation from the north generally dominates the air mass origin (Fig. S8a and d). However, during the day, the low-level thermal circulation becomes strong enough to influence the lowest trajectories, which start to cross the Po Valley in the second part of the day (Fig. S8b), in line with the simultaneous appearance of an aerosol layer in the ALC measurements. Together with their rotation during this day, the trajectories also decrease their altitude. At the end of the day, the air masses reaching the station travelled for more than 20 h on the surface of the Po basin. The end of the episode is marked again by northwestern provenance (Fig. S8c and f).

As in the first case, COSMO accurately predicts the advection of humid air at the same times as the ALC detects a thickening of the layer (Fig. b). At night, the simulated and measured RHs exceed 90 % at altitude and 80 % at ground level (Figs.  and S9b). Contrary to RH, SH does not show any daily cycle (Fig. S9b). A sudden SH increase (5 to 10 gkg-1) is clearly visible on 25 May at the time of the advection, while the values for the following days are almost constant except on the occasion of short showers (e.g. evening of 28 May).

The aerosol mass derived from the ALC is presented in Fig. c. The maximum concentration retrieved by this method within the aerosol layer is higher than 60 µgm-3 just before the formation of clouds at night. Again, FARM (Fig. d) qualitatively reproduces the afternoon increase in aerosol concentrations owing to transport from the boundaries; however the simulated concentrations are much lower (about 4–5 times) than the retrievals from the ALC and the advection arrival times are anticipated compared to the appearance of the thick layer from the ceilometer.

Hourly and sub-hourly PM10 surface concentrations (measured in Aosta–downtown and Aosta–Saint-Christophe and simulated by FARM) are presented in Fig. e. FARM correctly reproduces the morning rush-hour peak, but the concentrations are about half those from the PM samplers. The series in Fig. e shows an increase in PM10 surface concentration during 25 May, with persisting high values for the rest of the week, most noticeably during the night. Accordingly, PM daily means in Fig. a, b, g, h show a distinct increase in the whole region (the concentrations doubles) during the case study compared to the preceding and following days (and also compared to the 2015–2017 average concentrations for the same period, i.e. 12 µgm-3 for PM10 in Aosta–downtown and Donnas and 6 µgm-3 for PM2.5 in Aosta–downtown).

The number distribution and total particle number measured by the OPC in Aosta–Saint-Christophe are plotted in Fig. S9c and d, respectively. Figure S9 shows a notable increase in the number concentration during 25 May (from less than 200 to more than 800 particles cm-3) and in the afternoon of each day, and a decrease in the central part of each day as soon as the valley convection starts and the mixing layer height increases.

Percentage concentrations of nitrate, ammonium and sulfate are represented in Fig. c–e and account for about 20 %–25 % of the total PM10 mass (as a reference, this fraction represents less than 15 % for non-advection days in May–June 2017). Interestingly, relative nitrate concentration does not change much and does not reach the extreme values of the winter case study. Indeed, transfer of ammonium nitrate from particles to the gas phase, which is not measured, is favoured by higher temperatures . Conversely, ammonium and sulfate increase remarkably during advection, reaching typical concentrations of the Po Valley in that period . In particular, the sulfate concentration is much higher and more affected by the advection in May than during the winter case. The role reversal between nitrate and sulfate in case studies 2 and 3 results from the different sensitivity of those compounds to temperature and atmospheric conditions . Again, the contribution of inorganic species from the model does not agree with the analyses: the contribution by ammonium and sulfate is strongly underestimated, while the peaks in the simulated nitrate concentration are not reflected in the analyses.

As for the organic part, OM and EC are the main constituents of the remaining fraction, with about 60 % and 6 %, respectively. Although the available dataset is rather short, the OC/EC ratio during the event almost doubles (values of 6.1–7.2) compared to the value before (3.7) and after (3.3) the event. This increase in the OC fraction during transport episodes is confirmed by the long-term analysis .

The same morning–midday–afternoon modulation can be observed in the AOD from both the ALC and the sun photometer (Fig. S9e). The high AOD values in the first and last part of the day (up to 0.30–0.40 at 500 nm) and the decrease in the middle of the day (down to 0.12 at 500 nm) match the appearance of the layer seen by the ALC and its enhancement due to hygroscopic effects, in accordance with the results of for a typical site in the Po basin (in that case, for RH =90 %, the extinction coefficient increased on average to 180 % of the value measured for RH =0 %). The steep rise (from 0.8 to 1.7) in the Ångström exponent on 25 May (Fig. S9f) and its continuous increase in the following days (up to nearly 2.0) may be attributed to the advection of small particles to the measuring site. The SSA fluctuates around large values (generally between 0.9 and 1.0), typical of weakly light-absorbing or aged aerosol. Most interestingly, the volume distribution (Fig. S9g) exhibits an abrupt decrease in the peak diameter during the morning hours (e.g. from 0.5 µm at 06:40 UTC to 0.2 µm at midday on 27 May), strengthening the hypothesis of aerosol hydration at night and dehydration during the day as temperature increases and RH decreases.