Transport of Po Valley aerosol pollution to the northwestern Alps … · 2020. 7. 31. · tions to re-weight the emissions from outside the boundaries ... not least, local economy,

Atmos Chem Phys 19 10129ndash10160 2019httpsdoiorg105194acp-19-10129-2019copy Author(s) 2019 This work is distributed underthe Creative Commons Attribution 40 License

Transport of Po Valley aerosol pollution to the northwestern Alps ndashPart 2 Long-term impact on air qualityHenri Dieacutemoz1 Gian Paolo Gobbi2 Tiziana Magri1 Giordano Pession1 Sara Pittavino1 Ivan K F Tombolato1Monica Campanelli2 and Francesca Barnaba2

1ARPA Valle drsquoAosta Saint-Christophe Italy2Institute of Atmospheric Science and Climate ISAC-CNR Rome Italy

Correspondence Henri Dieacutemoz (hdiemozarpavdait)

Received 7 February 2019 ndash Discussion started 19 March 2019Revised 5 June 2019 ndash Accepted 26 June 2019 ndash Published 12 August 2019

Abstract This work evaluates the impact of trans-regionalaerosol transport from the Po basin on particulate matterlevels (PM10) and physico-chemical characteristics in thenorthwestern Alps To this purpose we exploited a multi-sensor multi-platform database over a 3-year period (2015ndash2017) accompanied by a series of numerical simulationsThe experimental setup included operational (247) verticallyresolved aerosol profiles by an automated lidar ceilometer(ALC) vertically integrated aerosol properties by a Sunskyphotometer and surface measurements of aerosol mass con-centration size distribution and chemical composition Thisexperimental set of observations was then complementedby modelling tools including numerical weather prediction(NWP) trajectory statistical (TSM) and chemical transport(CTM) models plus positive matrix factorisation (PMF) onboth the PM10 chemical speciation analyses and particle sizedistributions In a first companion study we showed and dis-cussed through detailed case studies the 4-D phenomenologyof recurrent episodes of aerosol transport from the pollutedPo basin to the northwestern Italian Alps Here we draw moregeneral and statistically significant conclusions on the fre-quency of occurrence of this phenomenon and on the quan-titative impact of this regular wind-driven aerosol-rich ldquoat-mospheric tiderdquo on PM10 air-quality levels in this alpine en-vironment Based on an original ALC-derived classificationwe found that an advected aerosol layer is observed at thereceptor site (Aosta) in 93 of days characterized by east-erly winds (ie from the Po basin) and that the longer thetime spent by air masses over the Po plain the higher thisprobability Frequency of these advected aerosol layers wasfound to be rather stable over the seasons with about 50 of

the days affected Duration of these advection events rangesfrom few hours up to several days while aerosol layer thick-ness ranges from 500 up to 4000 m Our results confirm thisphenomenon to be related to non-local emissions to act atthe regional scale and to largely impact both surface lev-els and column-integrated aerosol properties In Aosta PM10and aerosol optical depth (AOD) values increase respectivelyup to factors of 35 and 4 in dates under the Po Valley influ-ence Pollution transport events were also shown to modifythe mean chemical composition and typical size of particlesin the target region In fact increase in secondary speciesand mainly nitrate- and sulfate-rich components were foundto be effective proxies of the advections with the transportedaerosol responsible for at least 25 of the PM10 measuredin the urban site of Aosta and adding up to over 50 microgmminus3

during specific episodes thus exceeding alone the EU estab-lished daily limit From a modelling point of view our CTMsimulations performed over a full year showed that the modelis able to reproduce the phenomenon but markedly underes-timates its impact on PM10 levels As a sensitivity test weemployed the ALC-derived identification of aerosol advec-tions to re-weight the emissions from outside the boundariesof the regional domain in order to match the observed PM10field This simplified exercise indicated that an increase insuch ldquoexternalrdquo emissions by a factor of 4 in the model isneeded to halve the model PM10 maximum deviations andto significantly reduce the PM10 normalised mean bias fore-casts error (from minus35 to 5 )

Published by Copernicus Publications on behalf of the European Geosciences Union

10130 H Dieacutemoz et al Long-term impact on air quality

1 Introduction

Mountain regions are often considered pristine areas beingtypically far from large urban settlements and strong anthro-pogenic emission sources and thus relatively unaffected byremarkable pollution footprints However atmospheric trans-port of pollutants from the neighbouring foreland is not un-common owing to the synoptic and regional circulation pat-terns For example in several regions of the world thermallydriven flows represent a systematic characteristic of moun-tain weather and climate (Hann 1879 Thyer 1966 Kasten-deuch and Kaufmann 1997 Egger et al 2000 Ying et al2009 Serafin and Zardi 2010 Schmidli 2013 Zardi andWhiteman 2013 Wagner et al 2014a Giovannini et al2017 Schmidli et al 2018) favouring regular exchange of airmasses between the plain and the highland sites (Weissmannet al 2005 Gohm et al 2009 Cong et al 2015 Dhun-gel et al 2018) with likely consequences on human health(Loomis et al 2013 EEA 2015 WHO 2016) ecosystems(eg Carslaw et al 2010 EMEP 2016 Bourgeois et al2018 Burkhardt et al 2018 Allen et al 2019 Ambrosiniet al 2019 Rizzi et al 2019) climate (Ramanathan et al2001 Clerici and Meacutelin 2008 Philipona 2013 Pepin et al2015 Zeng et al 2015 Tudoroiu et al 2016 Samset 2018)and not least local economy through loss of tourism rev-enues (de Freitas 2003 Keiser et al 2018) These phenom-ena have worldwide relevance since nearly one-quarter ofthe Earthrsquos land mass can be classified as mountainous areas(Blyth 2002)

Among them the Alpine region (Fig 1a) is of particularinterest due to both its sensitive environment and its proxim-ity to the Po Valley (Tampieri et al 1981 Seibert et al 1998Wotawa et al 2000 Dosio et al 2002 Campana et al 2005Kaiser 2009 Finardi et al 2014) In fact the Po basin rep-resents a major pollution hotspot in Europe with large emis-sions from highly populated urban areas (about 40 of theItalian population lives in this region WMO 2012) intenseanthropogenic activities such as industry and agriculturecombined with a local topography promoting atmosphericstability (Chu et al 2003 Barnaba and Gobbi 2004 Schaapet al 2004 Van Donkelaar et al 2010 Bigi and Ghermandi2014 Fuzzi et al 2015 EMEP 2016 Belis et al 2017EEA 2017) Despite the efforts to decrease the number ofparticulate matter (PM) exceedances Italy is still failing tocomply with the European air-quality standards (EU Com-mission 2008 2018) the Po Valley being the major area re-sponsible for this situation

In a first companion paper (Dieacutemoz et al 2019) we in-vestigated the phenomenology of the aerosol-rich air massadvections from the Po basin to the northwestern Alps andspecifically to the Aosta Valley (Fig 1b) This was intro-duced and thoroughly described by means of three specifi-cally selected case studies each of them lasting several daysand monitored by a large set of instruments That investi-

gation evidenced clear features of this phenomenology andnotably

1 the aerosol transported to the northwestern Alps isclearly detectable and discernible from the locally pro-duced aerosol Detection of such polluted aerosol lay-ers over Aosta was primarily driven by remote-sensingprofiling measurements (automated lidar ceilometersALCs) these showing recurrent arrival of thick and el-evated aerosol-rich layers Good correspondence wasfound between the presence of these layers aloftchanges in column-averaged aerosol properties andvariations of PM surface concentrations and chemicalcomposition

2 the advected particles are small (accumulation mode)mainly of secondary origin weakly light-absorbing andhighly hygroscopic compared to the locally producedaerosol Clouds are frequently observed to form withinthese layers during the night

3 the air masses associated with the observed elevated lay-ers are found to originate from the Po basin and to betransported to the northwestern Alps by up-valley diur-nal wind systems and synoptic winds

4 the chemical transport model (FARM Flexible Airquality Regional Model) currently used by the local En-vironment Protection Agency (ARPA Valle drsquoAosta) isable to reproduce this transport from a qualitative pointof view and was usefully employed to interpret the ob-servations However it fails at quantitatively estimat-ing the PM mass contribution coming from outside theboundaries of the regional domain (approximately cor-responding to the Valle drsquoAosta region) likely due todeficiencies in the emission inventories therein and tounaccounted aerosol processes in the model (eg thoserelated to hygroscopicityaqueous-phase chemistry trig-gered by high relative humidities)

The present work analyses the same phenomenon but with along-term perspective The main aim of this study is to estab-lish the frequency of occurrence of aerosol transport from thePo Valley to the northwestern Alps and its impact on local airquality depending on the interplay between frequency andseverity of the episodes In particular this paper addressesthe following questions

1 How frequently does advection from the Po plain to thenorthwestern Alps occur What are the most commonmeteorological conditions favouring it

2 What are the average properties of the advected aerosolboth at the surface and in the layers aloft

3 How much does transport from the Po basin impactthe air-quality EU-regulated metrics in the northwesternAlps

Atmos Chem Phys 19 10129ndash10160 2019 wwwatmos-chem-physnet19101292019

H Dieacutemoz et al Long-term impact on air quality 10131

Figure 1 True colour corrected reflectance from the MODIS Terra satellite (httpworldviewearthdatanasagov last access 2 August 2019)on 17 March 2017 (a) Italy with indication of the Alpine and Po Valley regions of the Aosta Valley FARM regional domain (light blue rect-angle) and the COSMO-I2 domain (orange rectangle approximately corresponding to the boundaries of the national inventory) (b) Zoomover the Aosta Valley The circle markers represent the sites of (1) AostandashDowntown (2) AostandashSaint-Christophe (3) Donnas (4) Ivrea Athin aerosol layer over the Po basin starting to spread out into the Alpine valleys is visible in both figures

4 Can we effectively predict the arrival of aerosol-polluted air masses in the northwestern Alps How canwe improve our air-quality forecasts

To answer these questions we took advantage of long-term (2015ndash2017) and almost uninterrupted series of mea-surements carried out in the Aosta region The experimen-tal dataset thoroughly described by Dieacutemoz et al (2019)included measurements of vertically resolved aerosol pro-files by an ALC vertically integrated aerosol properties bya Sunsky photometer and surface measurements of theaerosol mass concentration size distribution and chemicalcomposition most of these series encompassing a time pe-riod of several years Indeed a great number of intensiveshort-term campaigns employing cutting-edge research in-struments and techniques were already performed in the Pobasin mainly focussing on its central eastern and southernparts (Nyeki et al 2002 Barnaba et al 2007 Ferrero et al2010 2014 Saarikoski et al 2012 Landi et al 2013 Dece-sari et al 2014 Costabile et al 2017 Bucci et al 2018Cugerone et al 2018) However continuous and multi-yeardatasets especially from ground-based stations are neces-sary to assess the influence of pollution transport on a longerterm (eg Meacutelin and Zibordi 2005 Clerici and Meacutelin 2008Kambezidis and Kaskaoutis 2008 Mazzola et al 2010Bigi and Ghermandi 2014 2016 Putaud et al 2014 Ar-vani et al 2016) In this context local environmental agen-cies operate stable networks for continuous monitoring ofair-quality and meteorological parameters and use standard-ised methodologies and universally recognised quality con-trol procedures

The study is organised as follows the investigated areais presented in Sect 2 together with the experimental setupand the modelling tools (Sect 3) Results are presented inSect 4 and include (a) a first original classification frame-

work based on ALC data and meteorological variables (b) along-term statistical analysis of the phenomenon and its im-pacts on surfacecolumn aerosol properties and (c) compar-ison between observations and CTM simulations Conclu-sions are drawn in Sect 5

2 Study region and experimental setup

We provide here a brief overview of the region of interest andof the experimental setup the latter being described with fulldetails in the companion paper (Dieacutemoz et al 2019)

The Aosta Valley located at the northwestern Italianborder (Fig 1a) is the smallest administrative region ofthe country being only 80 km by 40 km wide and hosting130 000 inhabitants The most remarkable geographical fea-ture of the region whose mean altitude exceeds 2000 m aslis the main valley approximately oriented in a SE-to-NWdirection and connecting the Po basin (about 300 m asl atthe southeastern side of the Aosta Valley region) to the MontBlanc chain (4810 m asl separating Italy and France) Sev-eral tributary valleys depart from the main valley The to-pography of the area triggers some of the most commonweather regimes in the mountains such as thermally drivenup-valley (daytime) and down-valley (nighttime) winds up-slope (daytime) and down-slope (nighttime) winds ldquoFoehnrdquowinds (Seibert 2012) from the west and frequent temper-ature inversions during wintertime anticyclonic days Quiteobviously those meteorological phenomena are responsiblefor most of the variability of the tropospheric constituents(eg pollutant concentrations Agnesod et al 2003 and wa-ter vapour Campanelli et al 2018)

The air-quality network of the regional EnvironmentalProtection Agency is designed trying to capture the to-pographical and meteorological heterogeneity of the in-

wwwatmos-chem-physnet19101292019 Atmos Chem Phys 19 10129ndash10160 2019

vestigated area In this study we mainly used data fromfour sites belonging to this network (Fig 1b) AostandashDowntown (580 m asl urban background) AostandashSaint-Christophe (560 m asl semi-rural) Donnas (316 m aslrural) and Ivrea (243 m asl urban background) The firsttwo measuring stations located 25 km apart are situated re-spectively in the centre and in the suburbs of Aosta (457 N74 E) the main urban settlement of the region (36 000 in-habitants) Car traffic domestic heating and a steel mill arethe main anthropogenic emission sources within the citywhose pollution levels are generally moderate (eg yearlyaverage PM10 concentration of about 20 microgmminus3 with sum-mertime and wintertime averages of 13 and 31 microgmminus3 re-spectively) Donnas is closer to the border with the Po basin(Fig 1b) and is expected to be strongly impacted by pollu-tion from outside the region In fact the PM10 concentrationin this rural site is about 18 microgmminus3 on a yearly average ieonly slightly lower than the concentration in Aosta We alsoconsider in our analysis the PM10 records collected in thecity of Ivrea a site in the Po Plain (31 microgmminus3 average PM10in 2017) located just outside the Aosta Valley in the ItalianPiedmont region (Fig 1b) This was done to check whetherand how much inaccuracies of the model in reproducing theaerosol loads over the Aosta Valley are due to difficulties insimulating the wind field in such a complex terrain In factthe city centre of Ivrea (24 000 inhabitants) is approximatelyin the middle between the measurement site (south of thecity) and the nearest cell of our model domain (north of thecity) the distance between these two points being only 2 kmThe altitude of the cell (from the digital elevation model usedin our simulations) is 241 m asl ie approximately the realone

AostandashDowntown hosts a monitoring station for contin-uous measurements of the aerosol concentration and com-position PM10 and PM25 daily concentrations are avail-able from two Opsis SM200 Particulate Monitor instru-ments An estimate of the PM10 hourly variability is further-more provided by a Tapered Element Oscillating Microbal-ance (TEOM) 1400a monitor (Patashnick and Rupprecht1991) but is not compensated for mass loss of semi-volatilecompounds (Green et al 2009) such as ammonium nitrate(eg Charron et al 2004 Rosati et al 2016) Moreoverthe collected PM10 samples are chemically analysed on adaily basis Ion chromatography (AQUIONICS-1000 mod-ules) is employed for determining the concentration of water-soluble anions and cations (Clminus NOminus3 SO2minus

4 Na+ NH+4 K+ Mg2+ Ca2+) based on the CENTR 162692011 guide-line Elementalorganic carbon (ECOC) using the thermalndashoptical transmission (TOT) method (Birch and Cary 1996)and following the EUSAAR-2 protocol (Cavalli et al 2010)according to the EN 169092017 are analysed alternativelyto metal concentrations (Cr Cu Fe Mn Ni Pb Zn As CdMo and Co) by means of inductively coupled plasma massspectrometry (ICP-MS) after acid mineralisation of the filterin aqueous solution (EN 149022005) In accordance with

this laboratory schedule (ie metal analyses on 2 consecu-tive days ECOC on the following 2 days followed by twosequences metalndashmetalndashECOC) over a 10 d period ECOCconcentrations are routinely provided on 4 d on average andmetal concentrations on 6 d Particle-bound polycyclic aro-matic hydrocarbons (PAHs) are detected continuously in realtime by a photoelectric aerosol sensor (EcoChem PAS-2000)Gas-phase pollutants such as NO and NO2 are also con-tinuously monitored and were used in the present work tohelp identification of urban combustion sources (mainly traf-fic and residential heating)

The atmospheric observatory in AostandashSaint-Christophe(WIGOS ID 0-380-5-1 Dieacutemoz et al 2011 2014a) is lo-cated on the terrace of the ARPA building At this site ad-vanced instrumentation is continuously operated checkeddaily and maintained Vertically resolved measurementsof aerosols and clouds are performed by a commerciallyavailable automated lidar ceilometer (ALC Lufft CHM15k-Nimbus) running since 2015 in the framework of the Ital-ian Alice-net (httpwwwalice-neteu last access 2 Au-gust 2019) and European E-PROFILE (httpse-profileeulast access 2 August 2019) networks The ALC emits intothe atmosphere 8 microJ laser impulses at a single wavelengthof 1064 nm with a frequency of 7 kHz and collects theirbackscatter by aerosol and molecules from the ground up toan altitude of 15 km with a vertical resolution of 15 m Thisreturn signal is averaged over 15 s and saved by the firmwareas range- overlap- and baseline-corrected raw counts (βraw)To convert the backscatter signal into SI units a Rayleighcalibration (Fernald 1984 Klett 1985 Wiegner and Geiszlig2012) is necessary This is accomplished based on data sam-pled during clear-sky nights This allows us to derive the par-ticle backscatter coefficient (βp) and the scattering ratio (SReg Zuev et al 2017) which is the primary ALC-derivedparameter used in this work to quantify the aerosol load It isdefined as

SR=βp+βm

βm (1)

βm being the molecular backscattering coefficient (thus SR=1 indicates a purely molecular atmosphere while SR in-creases with aerosol content)

At the same station a POM-02 Sunsky photometerhas operated since 2012 as part of the European ESR-SKYNET network (httpwwweuroskyradnet last access2 August 2019) The instrument collects the direct light com-ing from the Sun (every 1 min) to retrieve the aerosol opticaldepth (AOD τ(λ)) and its spectral dependence calculatedby fitting the Aringngstroumlm (1929) law to the measurements inthe 400ndash870 nm range

τ(λ)= bλminusa (2)

a being the Aringngstroumlm exponent Other aerosol optical andmicrophysical properties are retrieved from the light scat-tered from the sky in the almucantar plane (every 10 min)at 11 wavelengths (315ndash2200 nm) using the SKYRADpackversion 4 code The Sun photometer is calibrated in situ withthe improved Langley technique (Campanelli et al 2007)and was recently successfully compared to other referenceinstruments (Kazadzis et al 2018) Accurate cloud screen-ing is achieved using the Cloud Screening of Sky Radiome-ter data (CSSR) algorithm by Khatri and Takamura (2009)Special care is needed to retrieve among the other param-eters the aerosol single scattering albedo (SSA) In fact itis known that the SSA retrieval by the SKYRADpack ver-sion 4 is problematic and sometimes unnaturally close tounity (Hashimoto et al 2012) independently of the AODAlthough improvements are expected in the next versionsof SKYRADpack data collected within the European ESR-SKYNET network are still processed with version 4 There-fore only SSAs lower than 099 at all wavelengths were re-tained in our analysis

Finally a Fidas 200s Optical Particle Counter (OPC) isused in AostandashSaint-Christophe to yield an accurate estima-tion of the aerosol size distribution (018ndash18 microm) in the prox-imity of the surface (Pletscher et al 2016) This instrumentalthough not directly measuring the aerosol mass obtainedthe certificate of equivalence to the gravimetric method byTUumlV Rheinland Energy GmbH on the basis of a laboratorytest and a field test A PM10 comparison with the gravimet-ric technique was additionally performed in AostandashSaint-Christophe and provided satisfactory results (29 d slope108plusmn004 interceptminus38plusmn14 microg mminus3R2

= 096) Furtherinstruments to monitor trace gases (Dieacutemoz et al 2014b) areemployed at the station and the corresponding datasets willbe investigated in future studies

Some of the air-quality parameters monitored in AostandashDowntown are also measured at the Donnas station ie PM10daily concentrations (Opsis SM200) and nitrogen oxides(Horiba APNA-370 chemiluminescence monitor) Finallythe three stations are equipped with instruments for trackingthe standard meteorological parameters such as temperaturepressure relative humidity (RH) and surface wind velocity

The station of Ivrea features among other instruments aTCR Tecora CharlieSentinel PM10 sampler PM10 concen-trations are then determined by a gravimetric technique

A list of all measurements with relevant operating periodand subset considered in this study is presented in Table 1

3 Modelling tools

Numerical models and statistical techniques were used to in-terpret and complement the observations described above Anumerical weather prediction model (COSMO Consortiumfor Small-scale Modeling httpwwwcosmo-modelorg lastaccess 2 August 2019) was employed to drive a chemi-

cal transport model (FARM Flexible Air quality RegionalModel) and a Lagrangian model (LAGRANTO) These toolswere thoroughly described by Dieacutemoz et al (2019) andare only briefly recalled here (Sect 31) Trajectory statisti-cal models (TSMs) and positive matrix factorisation (PMF)adopted to interpret the long-term series of measurementsused in this work are fully described in Sects 32 and 33respectively

31 Numerical atmospheric models

COSMO is a non-hydrostatic fully compressible atmo-spheric prediction model working on the meso-β and meso-γscales (Baldauf et al 2011) The forecasts inclusive of thecomplete set of parameters (such as the 3-D wind velocityused here) for eight time steps (from 0000 to 2100 UTC)are disseminated daily in two different configurations bythe meteorological operative centrendashair force meteorologi-cal service (COMET) a lower-resolution version (COSMO-ME 7 km horizontal grid and 45 levels vertical grid 72 hintegration) covering central and southern Europe and anudged higher-resolution version (COSMO-I2 or COSMO-IT 28 km 65 vertical levels 2 runs dminus1) covering Italy (or-ange rectangle in Fig 1a) Owing to the complex topographyof the Aosta Valley and the consequent need to resolve asmuch as possible the atmospheric circulation at small spa-tial scales we used the latter version in the present work(cf Sect 45 for a discussion of possible effects of the finitemodel resolution in complex terrain)

FARM version 47 (Gariazzo et al 2007 Silibello et al2008 Cesaroni et al 2013 Calori et al 2014) is a four-dimensional Eulerian model for simulating the transportchemical conversion and deposition of atmospheric pollu-tants with a 1 km spatial grid 1 h temporal resolution and16 different vertical levels (from the surface to 9290 m)FARM can be easily interfaced to most available diagnosticor prognostic numerical weather prediction (NWP) modelsas done with COSMO in the present work Pollutant emis-sions from both area and point sources can be simulatedby FARM taking into account transformation of chemicalspecies by gas-phase chemistry and dry and wet removalParticularly interesting for our study is the aerosol mod-ule (AERO3_NEW) coupled with the gas-phase chemicalmodel and treating primary and secondary particle dynamicsand their interactions with gas-phase species thus account-ing for nucleation condensational growth and coagulation(Binkowski 1999) Three particle size modes are simulatedindependently the Aitken mode (diameter D lt 01 microm) theaccumulation mode (01 microm ltD lt 25 microm) and the coarsemode (D gt 25 microm) FARM is only run over a small domain(light-blue rectangle in Fig 1a b) roughly correspondingto the Aosta Valley A regional emission inventory (ldquolocalsourcesrdquo updated to 2015) is supplied to the CTM over thesame area to accurately assess the magnitude of the pollu-tion load and its variability in time and space Data from

Table 1 Observation sites measurements and instruments employed in this study

Station Elevation(m asl)

Measurement Instrument Dataavailability

Used in thisstudy

AostandashDowntown 580 PM10 hourly concentration TEOM 1400a 1997ndashnow 2015ndash2017

PM10 and PM25 dailyconcentrations

Opsis SM200 2011ndashnow 2015ndash2017

Water-soluble anioncationanalyses on PM10 samples

Dionex Ion ChromatographySystem

2017ndashnow 2017

ECOC analyses on PM10samples

Sunset thermo-optical anal-yser

2017ndashnowa 2017

Metals on PM10 samples Varian 820 MS 2000ndashnowb 2017

Total PAHs EcoChem PAS-2000 1995ndashnow 2017

NO and NO2 Horiba APNA-370 1995ndashnow 2017

Standard meteorologicalparameters

Vaisala WA15 (wind) 1995ndashnow 2015ndash2017

AostandashSaint-Christophe(ARPAobservatory)

560 Vertical profile of attenuatedbackscatter and derivedproducts

CHM15k-Nimbus ceilometer 2015ndashnow 2015ndash2017

Aerosol columnar properties POM-02 Sunsky radiometer 2012ndashnowc 2015ndash2017

Surface particle sizedistribution

Fidas 200s optical particlecounter

2016ndashnow 2016ndash2017

AostandashSaint-Christophe(weather station)

545 Standard meteorologicalparameters

Micros (wind) 1974ndashnow 2015ndash2017

Donnas 316 PM10 daily concentration Opsis SM200 2010ndashnow 2015ndash2017

NO and NO2 Teledyne API200E 1995ndashnow 2015ndash2017

Ivrea 243 PM10 daily concentration TCR Tecora CharlieSentinel PM

2006ndashnow 2017

a The analysis is performed on 4 out of 10 d according to the laboratory schedule b The analysis is performed on 6 out of 10 d according to the laboratory schedulec Underwent major maintenance in the second half of 2016 and January 2017

a national inventory and CTM (QualeAria here referred toas ldquoboundary conditionsrdquo outer rectangle in Fig 1a) takenalong the border of the inner (light blue) rectangle are alsoused to estimate the mass exchange from outside the bordersof the FARM domain

32 Trajectory statistical models

The forecasted wind velocity profile from COSMO was hereused as an input parameter into the publicly available LA-GRANTO Lagrangian analysis tool version 20 (Sprengerand Wernli 2015) to numerically integrate the 3-D windfields and to determine the origin of the air masses sam-pled by the ALC over AostandashSaint-Christophe We set upthe program to start an ensemble of 8 trajectories in a cir-cle of 1 km around the observing site and at seven different

altitudes from the ground to 4000 m asl for a total of 56trajectories per run A backward run time of 48 h was con-sidered sufficient on average to cover most of the domainof the meteorological model while minimising the numericalerrors

Back-trajectories calculated with LAGRANTO were thenemployed in TSMs to provide a general picture of the geo-graphical distribution of the probable aerosol sources In thisbroadly used technique the NWP domain is divided into gridcells (i and j indices) and air parcels arriving at a specific re-ceptor site are analysed When a cell ij is crossed by a back-trajectory l the tracer concentration cl measured at the arrivalpoint of that trajectory (receptor) is considered Finally foreach cell a weighted average Pij is calculated as follows to

yield a map of the possible source areas (eg Kabashnikovet al 2011)

sumLl=1F(cl)τij (l)sumL

l=1τij (l) (3)

where L is the total number of trajectories and τij is thetime spent by the trajectory l in the grid cell ij F(cl) is afunction of the concentration at the receptor and depend-ing on the chosen technique can be the concentration itself(concentration-weighted trajectory (CWT) method eg Hsuet al 2003) the logarithm of the concentration (concentra-tion field (CF) method Seibert et al 1994) or the Heavisidestep function H(clminuscT) where cT is a concentration thresh-old (potential source contribution function (PSCF) methodeg Ashbaugh et al 1985) generally the 75th percentile ofthe concentration series In this last case Pij can be statisti-cally interpreted as the conditional probability that concen-trations above the threshold cT at the receptor site are relatedto the passage of the relative back-trajectory through the lo-cation ij (eg Squizzato and Masiol 2015) Additional it-erative methods exist to reduce trailing effects and to betteridentify pollution hotspots (eg Stohl 1996) Moreover theobtained field Pij can be further weighted as a function ofthe number of trajectories passing through each cell thus re-ducing the impact of cells with limited statistics (Zeng andHopke 1989)

Any series of atmospheric measurements can be used asthe concentration variable cl in Eq (3) In this work weachieved especially meaningful results employing the scoresfrom the PMF decomposition (Sect 33) ie the contribu-tions of the identified sources to the PM10 daily concen-tration measured at AostandashDowntown The same approachcoupling PMF and TSMs was employed in other recent stud-ies (eg Bressi et al 2014 Waked et al 2014 Zong et al2018) However since only daily average information isavailable from the chemical speciation the PMF output wasrepeated eight times per day in order to allow correspondenceto the eight back-trajectories per day issued by COSMO andLAGRANTO We present here results obtained with CF hav-ing checked that application of CWT and PSCF methods pro-vides similar outcomes The COSMO-I2 domain was thus di-vided into 130times90 grid cells and 48 h back-trajectories wereconsidered Only points of the back-trajectories at altitudeslower than 2000 m asl were taken into account this beingthe approximate average height of the mixing layer in the Pobasin (Dieacutemoz et al 2019) Similarly since the arriving airparcels should be able to impact surface PM levels only tra-jectories ending close to the surface over Aosta were exam-ined (ie altitudes lt 1500 m asl the model surface altitudein the Aosta area being about 900 m asl) To reduce statis-tical noise every value Pij of the resulting map was thenmultiplied by a weighting factor wij linearly varying from0 (for Nij le 20 end points in a cell) to 1 (for Nij ge 200)

ie wij =min1 max0Nijminus20180 To provide an idea of the

effect of this weighting procedure TSM maps without anyweighting (ie wij = 1) have been included in the Supple-ment (Fig S7)

33 Positive matrix factorisation

The positive matrix factorisation (PMF Paatero and Tapper1994 Paatero 1997) technique as implemented in the USEPA PMF50 software (Norris and Duvall 2014) was em-ployed to study the 2017 series of aerosol chemical analy-ses and air-quality measurements in AostandashDowntown Themethod allowed us to identify the possible emission sourcesof the observed aerosol and notably to discriminate its localand non-local origin PMF requires a long multivariate seriesof chemical analyses matrix X (ntimesm the n rows being thesamples and m columns representing the chemical species)and decomposes it into the product of two positive-definitematrices ie G (ntimesp matrix of factor contributions p be-ing the number of factors chosen for the decomposition) andF (ptimesm matrix of factor profiles) plus residuals (matrix E)

X=GF+E (4)

A solution is found by minimizing the so-called objec-tive function Q ie the squared sum of E weighted by themeasurement uncertainties A total of 100 runs were per-formed in this work for each dataset to find an optimal so-lution Here we considered three different datasets X (a) theoverall dataset of anioncation analyses (data available al-most every day in 2017 n= 360 ldquoPMF dataset ardquo) (b) asubset of dataset (a) selecting those measurements also hav-ing coincident ECOC analyses (n= 132 ldquoPMF dataset brdquo)and (c) a subset of (a) selecting those measurements alsohaving coincident metal speciation (n= 209 ldquoPMF datasetcrdquo) To make the decomposition of the three series compa-rable an ldquounidentifiedrdquo contribution corresponding to thecarbonaceous species only included in PMF dataset b wasadded to PMF datasets a and c This was done in the fol-lowing way we first estimated the total mass of crustal el-ements using the measured water-soluble calcium concen-tration as a proxy with an empirical conversion factor of10 (eg Waked et al 2014 note that a more rigorous es-timate for the soil component was obtained from the PMFanalysis itself and confirmed this factor) we then subtractedthe sum of the available chemical components (including theestimated mass from soil components) from the total mea-sured PM10 concentration thus closing the mass balanceThis difference was then added in PMF datasets a and c asthe ldquounidentifiedrdquo source

Finally NO NO2 and total PAHs measured at the samesite (AostandashDowntown) were included in the PMF to helpidentification of local pollution sources The number of fac-tors for each dataset was chosen based on physical inter-pretability of the resulting factors (Sect S4) and on the

QQexp ratio The latter is the ratio between the objectivefunction obtained with the selected number of factors (Qintroduced before) and its expected value (Qexp) Elevated(ie gt 2) QQexp ratios could indicate that some samplesandor species are not well modelled and could be better ex-plained by adding another source (Norris and Duvall 2014)The measured PM10 concentration was selected as the totalvariable to determine the contribution of each mode to thePM10 concentration

PMF was additionally performed on volume size distribu-tions measured by the Fidas OPC in AostandashSaint-Christophe(Sect 433) the m columns representing particle sizes in-stead of chemical species

4 Results

41 Daily classification schemes based on ALCmeasurements or meteorology

Information on the vertical dimension obtained by the ALCwas a key factor in identifying the events of aerosol pollu-tion transport over the northwestern Alps (eg Dieacutemoz et al2019) Therefore a first step of our analysis was to set upa classification scheme of those advection dates based onALC measurements To this purpose we used the longestALC record available (2015ndash2017) and coupled it to mete-orological measurementsforecasts Since to our knowledgeno previous ALC-based classification method is available inthe scientific literature we defined an original daily resolvedclassification scheme based on ceilometer measurements togroup dates according to specific ALC-observed conditionsA graphical overview with examples for each class identifiedis given in Fig 2 Overall we identified six classes (AndashF) ofdays

ndash days ldquoArdquo no aerosol layer or only low (ie lt 500 mfrom the ground) aerosol layers visible for the wholeday These days are intended to represent unpollutedconditions or cases only affected by local sources sinceaerosol transport from remote regions generally mani-fests as more elevated layers as found by Dieacutemoz et al(2019)

ndash days ldquoBrdquo only brief episodes of layers developing fromthe surface to altitudes gt 500 m agl observed duringthe 24 h The origin of the aerosol layers of this inter-mediate class is uncertain since they could either resultfrom weak (not completely developed) advections fromoutside the investigated region or from puffs of locallyproduced aerosol For this reason data in class ldquoBrdquo wereused only as a transitional level but not to draw defini-tive conclusions

ndash days ldquoCrdquo detection of an elevated well-developedaerosol layer (usually in the afternoon) and persistentduring the night

Figure 2 Example of ALC images representative of each category(AndashF) described in Sect 41 The corresponding dates are 1 Decem-ber 2015 19 October 2016 20 April 2016 11 April 2015 1 Novem-ber 2017 and 27 January 2017 The dashed lines for categories CndashFidentify the sigmoid interpolation to the SR= 3 envelope using theautomated algorithm as explained in Sect 42

ndash days ldquoDrdquo detection of the aerosol layer from the previ-ous day dissolving during the morning hours No layersin the afternoon

ndash days ldquoErdquo detection of a first aerosol layer from the pre-vious day dissolving (or strongly reducing) in the morn-ing and appearance of a separated structure in the after-noon

Figure 3 Frequency distribution of the aerosol layers observed inAostandashSaint-Christophe based on the ALC classification describedin Sect 41 A no layer B short episodes C layer detected in theafternoon D leaving layer E layer dissolving in the morning anda separated structure in the afternoon and F persistent layer

ndash days ldquoFrdquo thick aerosol layer persisting all day long

Note that although we also developed automatic recogni-tion algorithms for classification purposes we finally chosea classification based on visual inspection (for a total of 928daily images over the 3-year period) to ensure the maximumquality of the flagging and avoid further sources of error

The ALC classification described above was used to de-rive the seasonal frequency distribution of each aerosol ad-vection class Relevant results are shown in Fig 3 Days notfalling into those classes (eg complex-shaped layers pres-ence of low clouds or heavy rain hiding the aerosol layer)or including other kinds of aerosol layers (eg Saharan dust)were flagged as ldquoOtherrdquo and were not considered for furtheranalyses (this accounting for a maximum 26 of days insummer) The results reveal that clearly detected aerosol ad-vections (cases C to F) occur half (50 ) of the days in sum-mer and autumn and with a slightly lower frequency (45 )in winter and spring A higher percentage of clear days (A)occurs in winter and spring persistent layers (F) are mostlyfound in winter while cases E are more frequent in summer

To assess how the ALC classification described aboveis related to the circulation patterns we used the surfacemeteorological observations in AostandashSaint-Christophe asdrivers of a second independent classification scheme assummarised in Table 2 Seasonal frequency distributionsfrom this meteo-based scheme are displayed in Fig 4 Theseshow that weather circulation types are remarkably variableduring the year and help in interpreting the ALC-based re-sults (Fig 3) In fact winter is mostly characterised by windcalm (53 of the days) which sometimes favours stagnationand persistent aerosol layers (F) Plain-to-mountain diurnalwinds gradually increase from winter (only 11 of the daysin that season which reflects the highest percentage of clear

days (A)) to summer (68 of the days) when the thermallydriven regional circulation represents the main mechanismcontributing to transport of polluted air masses (E) Foehnwinds are fundamental processes especially in spring (22 of the days) leading to the removal of pollutants and fre-quent occurrence of clear days (A) in that season Finallychannelled synoptic flows from the east (or from the west)are also partly responsible for pushing polluted air massestowards (or away from) the Aosta Valley although they aremarkedly less frequent compared to the thermal winds mech-anism

The association between the ALC- and meteo-based clas-sification schemes was explored using contingency tables(Fig 5a) To this aim the ALC classification (AndashF) was fur-ther simplified and reduced to two main cases presence ofan arriving or stationary aerosol layer (classes C E and F)and absence of any thick aerosol layer (class A) Classes B(short and uncertain episodes) and D (layer leaving duringthe morning) were not used for the next considerations toavoid confounding conditions Figure 5a shows that the pres-ence of a thick aerosol layer is strongly connected to the winddirection as expected In fact this scheme reveals that thelayer appearance can be easily explained using the wind di-rection as the only information when air masses come fromthe east the probability of detecting an aerosol layer overthe Aosta Valley is 93 confirming the Po basin as a heavyaerosol source for the neighbouring regions independentlyof the day and period of the year Conversely this probabil-ity reduces to only 2 of the cases when the wind blowsfrom the west In the period addressed we thus detected fewexceptions to the perfect correspondence (100 and 0 )which can however still be explained For cases of easterlywinds and no aerosol layer found (7 ) possible reasons are

ndash the diurnal wind although clearly detected by the sur-face network was of too short a duration or too weakto transport the polluted air masses from the Po basinto AostandashSaint-Christophe (at least sim 40 km must betravelled by the air masses along the main valley) Inour record this was for example the case of 14 and18 September 2015 and 17 June 2016

ndash back-trajectories were indeed channelled along the cen-tral valley however they did not come from the Pobasin but rather from the other side of the Alps Thiswas the case for instance of 20 August 2015 in whichair masses came from the Divedro Valley (northeast ofthe Aosta Valley) after crossing the Simplon pass (be-tween Switzerland and Italy)

These exceptions also help understand why in summerabout 70 of the days feature breeze systems (Fig 4) whileaerosol layers are detected on only about 50 of the days(Fig 3) Indeed some part (7 Fig 5a) of this 20 dis-crepancy can be attributed to the above events the remain-ing 13 being likely associated with days with a complex

Table 2 Classification of weather regimes used in the study based on wind speed (|v|) wind direction daily duration of the condition (1t)and other measured meteorological variables

Primary Secondary Definitionclassification classification

Easterly winds Channelled synoptic flow from the east |v|gt 1 msminus1 1t gt12 h wind direction 0ndash180

Night (1900ndash0800 UTC)a calm wind (|v|lt 1 msminus1)Diurnal wind systems (east) or low westerly wind (|v|lt 15 msminus1)

Day (0800ndash1900 UTC)a easterly wind (|v|gt 15 msminus1) 1t gt4 h

Westerly winds Foehn (west) |v|gt 4msminus1 1t gt 4 h wind direction lt 25 or gt 225b RHlt 40

Channelled synoptic flow from the west no foehn |v|gt 1 msminus1 1t gt12 h wind direction 180ndash360

Wind calm ndash |v|lt 1 msminus1 1t gt20 h

Other Precipitation Accumulated daily precipitation gt 1mm 1t gt4 h

Unclassified Every day that does not fall into the previous categories

a Both conditions must be met in order to discriminate such diurnal wind systems (inactive at night) from synoptic winds b The directional range for the foehn cases wasdetermined experimentally by comparison with manual classification from a trained weather observer

Figure 4 Frequency distribution of circulation conditions in AostandashSaint-Christophe based on the meteorological classification describedin Table 2 Easterly winds (diurnal wind systems and channelled synoptic flow from the east) are represented as orange slices wind calmconditions in yellow and westerly winds (foehn and channelled synoptic flow from the west) in blue Precipitation and unclassified days arerepresented as grey slices and are not used further in the study

aerosol structure (not classified in Fig 2) or with days af-fected by low clouds andor desert dust events These dayswere therefore labelled as ldquoOtherrdquo in Fig 3

There are few cases (2 Fig 5a) in our record in whichconversely an aerosol layer was associated with westerlyrather than easterly winds These are as follows

ndash on 5 February and 15 March 2016 the prevalent windwas from the west and the days were correctly iden-tified as ldquolsquoChannelled synoptic flow from the westrdquo

and ldquoFoehnrdquo respectively However as soon as thewind turned and the valleyndashmountain diurnal circulationstarted the aerosol layer arrived

ndash on 24 March 2016 a real case of transbound-arytransalpine advection occurred and an aerosol-richair mass was transported from the polluted French val-leys on the other side of the Mont Blanc chain tothe Aosta Valley Although heavy pollution episodescan occur also on the French side of the Alps (eg

Figure 5 (a) Contingency table showing occurrences of the aerosol layer seen by the ALC and wind direction The blue boxes representcases when the aerosol layer is not visible (day types A) and pink boxes cases when an arriving or stationary aerosol layer is revealed bythe ALC (day types C E and F) The numbers inside the boxes refer to the frequency and the number of occurrences for each couple ofwindndashlayer classes (b) Occurrence of the aerosol layer seen by the ALC and residence time of 48 h back-trajectories in the Po plain

Chazette et al 2005 Brulfert et al 2006 Bonvalotet al 2016 Chemel et al 2016 Largeron and Staquet2016 Sabatier et al 2018) it is very rare to clearlyspot an aerosol layer transported from that region toAosta since in those cases air masses coming from thewest must have crossed the Alps and are almost alwaysmixed with clean air from the uppermost layers there-fore only a dim aerosol backscatter signal is usually no-ticed from the ALC

Overall the good correspondence between the measuredwind regimes and the aerosol layer detection by the ALC(Fig 5a) suggests that the same association could be em-ployed to forecast the arrival of an advected aerosol layerbased on the predicted wind fields As an example of thesimple forecasting capabilities of this phenomenon Fig 5billustrates a contingency table relating the occurrence of anaerosol layer in AostandashSaint-Christophe to the residence timeover the Po Valley of the 48 h back-trajectories arriving at theAostandashSaint-Christophe observing site Again a direct clearconnection between both variables can be seen with a 100 correspondence for residence times of air masses over the Pobasin gt 10 h

Finally it is worth mentioning that the proven high corre-lation between the circulation patterns (observed andor sim-ulated) and the presence of advected polluted layers in theAosta area may be exploited in the future to reconstruct theimpacts of aerosol transport on air quality over the Alpine re-gion back to previous periods not covered by the ALC mea-surements (a 40-year long meteorological database is avail-able in the region)

42 Vertical and temporal characteristics of theadvected aerosol layer

The 2015ndash2017 ALC time series in AostandashSaint-Christopheis summarised in Fig 6 showing the 1 h resolved monthlyaverage SR daily evolution A data screening was prelimi-narily performed by removing cloud periods with less than30 min measurements per hour and points lt 1 week of mea-surements per month The plot clearly shows that the max-imum aerosol backscatter is generally found in the earlymorning and in the late afternoon likely due to couplingbetween aerosol transport and hygroscopic effects (Dieacutemozet al 2019) and not as usual for plain sites in corre-spondence to the midday development of the mixing bound-ary layer Also the altitude of the layer varies with seasonIncidentally months affected by exceptional events can besharply distinguished the frequent Saharan dust events inJune and August 2017 and the forest fire plumes from Pied-mont in October 2017 the latter again affecting the Aosta sitein the afternoon because of the thermally driven circulationA similar climatology showing the results in terms of aver-age PM concentration profiles retrieved by the ALC is givenin Fig S1 of the Supplement In that case the conversionfrom backscatter to PM10 was obtained using the methodol-ogy described by Dionisi et al (2018)

To quantitatively explore the spatial and temporal char-acteristics of the ALC-detected aerosol layer over the longterm we developed an automated procedure described in de-tail in the Supplement (Sect S2) to identify and fit with asigmoid curve the spacendashtime region of the ALC profiles af-fected by the aerosol advection (using SRge 3 as a thresholdvalue) This allowed us to objectively determine for exam-ple the height and the duration of these advections The long-term statistics of these two variables is summarised in Fig 7

Figure 6 Monthly averages of the SR daily evolution (hourly measurements from 0000 to 2400 UTC) from the ALC in AostandashSaint-Christophe (2015ndash2017) Although ALC measurements extend up to 15 km altitude we limited the figure to 5000 m agl to better show thelowermost levels where aerosol transport from the Po basin occurs (Dieacutemoz et al 2019) Points with insufficient statistics are not plotted(white areas cf main text)

Figure 7 (a) Aerosol layer top altitude for the classified daysMarker colour represents the type of the advection while the markershape represents the time of the day morning (am) afternoon(pm) or all day (applicable to cases F only) (b) Overall durationof the aerosol layer in a day (morning and afternoon durations weresummed up in case E) The boxplots on the right represent syntheticinformation for each day type (same y scale as the relative charts onthe left) Missing measurements in summer 2015 are due to the re-placement of the ALC laser module

The altitude of the polluted aerosol layer (Fig 7a) displaysas expected a clear seasonal cycle with minimum in winterand maximum up to 4000 m agl in summer The overallcycle mimics the variability of the convective boundary layer(CBL) height found in the Po basin by other studies (Barnabaet al 2010 Decesari et al 2014 Arvani et al 2016) withsomewhat higher values that reflect the modification of theboundary layer structure in mountainous areas (De Wekkerand Kossmann 2015 Serafin et al 2018) Notably strongermulti-scale thermally driven flows (eg the ones developingon the slopes of the valley) further redistribute the aerosolparticles in the vertical direction thus pushing the upper limitof the CBL to the ridge height Average values of the differ-ent classes represented as boxplots in Fig 7 are clearly im-pacted by the season of their maximum frequency over theyear (Fig 3) In fact minimum altitude of the layer top wasfound on days F owing to their maximum occurrence in win-ter while the top altitude maximum was found on days Ebeing these mostly summertime events Great variability wasalso found for the duration of the advection (Fig 7b) rangingfrom a few hours in the weakest events to 24 h per day in theextreme cases (full day by definition for cases F) The cou-pling between the duration of the events and their absoluteaerosol load determines the impact of the investigated phe-nomenon on surface air quality as discussed in the followingparagraphs

43 Impact of Po Valley advections on northwesternAlps surface-level PM loads and physico-chemicalcharacteristics

To quantify the impact of the aerosol layers detected by theALC on the air quality at ground level we first investigated

how the PM concentration daily evolution measured by thesurface network changes depending on the ALC-identifiedclasses (Sect 431) We then used the positive matrix fac-torisation of the multivariate dataset of chemical analyses toexplore the chemical markers associated with the transportfrom the Po basin (Sect 432) The effectiveness of the iden-tified markers and the coupling between chemistry and me-teorology were also confirmed by the use of trajectory statis-tical models Furthermore a link between aerosol chemicaland physical properties was investigated in Sect 433 whereparticle chemical composition is coupled to measurementsof aerosol particle size Finally a preliminary assessment ofthe effect of the investigated advections on surface pollutantsother than particulate matter (eg NOx partitioning) was alsoperformed and is reported in the Supplement (Sect S5)

431 Surface PM variability in relation to the ALCprofilesrsquo classification scheme

We started investigating the effect of aerosol transport onthe Aosta Valley surface air quality by analysing the varia-tions of the daily mean PM concentrations measured by theARPA surface network as a function of the ALC classes in-troduced in Sect 41 Figure 8a shows the relevant resultsresolved by season It is quite evident that PM10 concentra-tions in AostandashDowntown are in fact highly correlated withthe presencestrength of the aerosol layers seen by the ALCPM10 on days with a persistent aerosol layer (class F) wasfound to be 2 to 35 times higher on average than on non-event days (class A this difference being statistically signif-icant for all seasons as proven by the p value lt 005 fromthe Mann and Whitney 1947 test) Similar behaviour wasseen in PM25 concentrations measured at the same station(Fig S2a) and in the PM25PM10 ratio (Fig S2b) the latterindicating that aerosol particles are smaller on average ondays when the ALC detects a thick aerosol layer comparedto non-event days Causality between the presence of an el-evated layer and variations of the PM surface concentrationhowever cannot be rigorously inferred at this point since inprinciple common environmental conditions affecting bothPM at the ground and the aerosol in the layers aloft couldgenerate the observed correlation Nevertheless it is inter-esting to note that this effect is less detectable for other pol-lutants measured by the network In fact concentrations oftypical locally produced pollutants such as NOx and PAHs(Fig 8b and c) do not change as much as PM among theALC classes with only rare exceptions (eg class A which isslightly influenced by foehn winds and class F in which tem-perature inversionslow mixing layer heights may enhancethe effect of local surface emissions) These results indicatethat the correlation (Fig 8a) does not trivially originate fromcommon weather conditions or from the uneven distributionof the ALC classes among the seasons

Furthermore large correlation was also found between theALC classification only available in AostandashSaint-Christophe

Figure 8 Daily averaged surface concentrations (2015ndash2017) ofPM10 (a) nitrogen oxides (b) and polycyclic aromatic hydrocar-bons (c) in AostandashDowntown as a function of the ALC classes andseason (d) PM10 surface concentration at the Donnas station TheEU-established PM10 daily limit of 50 microg mminus3 and the NO2 hourlylimit of 200 microg mminus3 are also drawn as references (horizontal dashedlines)

and the surface PM10 concentration recorded in Donnas(Fig 8d) Since the two stations are located at 40 km dis-tance this result suggests that a major role is played by large-scale dynamics rather than by local sources and that the phe-nomena under consideration have at least a regional-scale ex-tent As a further test we correlated the daily PM10 concen-trations measured in Donnas with those measured in AostaTo this purpose we considered the anomaly (1PM10 Eq 5)with respect to the monthly moving average (〈PM10〉m) inorder to avoid large fictitious correlations due to the sameyearly cycle (maximum PM concentration in winter and min-imum in summer) and only focus on the short-term dynam-ics

1PM10(t)= PM10(t)minus〈PM10〉m(t) (5)

where t is a specific day The scatterplot of these anoma-lies is shown in Fig 9 It highlights the good correlation be-tween the 1PM10 at the two sites (ρ = 073 when consider-ing all classes and ρ = 076 when only considering advectioncases ie classes C to F) Clustering of these results usingthe ALC classification (circle colours) further shows the dif-ferent 1PM10 associated with the different classes and thegood correspondence of this effect at both sites Notably inthe most severe cases (E and F) the anomalies in Donnas aregenerally larger than in AostandashDowntown (the best fit linebeing y = 11x+ 08 for cases E and F only) which is com-patible with this site being closer to the Po basin (Fig 1)

The advected aerosol layers were also found to impact thedaily evolution of surface PM concentrations To investigatethis aspect we used hourly and sub-hourly PM10 data fromboth the Fidas 200s OPC in AostandashSaint-Christophe andthe TEOM in AostandashDowntown Figure 10 shows the PM10daily cycles for cases A (no advection) C and D for bothAostandashSaint-Christophe (Fig 10a) and AostandashDowntown(Fig 10b) Despite the instrumentsrsquo limitations described inSect 2 the figure shows that local effects play an impor-tant role in the PM10 daily evolution as visible from themorning and afternoon peaks In fact these maxima com-mon to most pollutants are the results of the coupled ef-fect of varying local emissions and boundary layer heightduring the day The daily cycle of PAH concentrations isadded to the plot as a proxy of the diurnal cycle from lo-cal sources (dashed line right y axis) This cycle is simi-lar to the PM10 one for case A Conversely the influenceof the advections (C and D) is clearly visible in the corre-sponding PM10 cycles In fact Fig 10 shows the PM10 con-centration to markedly increase in the afternoon of days C(by 10 and 07 microg mminus3 hminus1 in AostandashSaint-Christophe andAostandashDowntown respectively) and to markedly decrease ondays D (byminus13 andminus07 microg mminus3 hminus1) This effect is similarat the two sites although less marked in AostandashDowntownespecially at night This is probably due to TEOM loss ofvolatile species in the advected aerosol (see also the PMF

results on chemical analysis in the following section) Simi-lar results (not shown) were obtained when splitting the dataon a seasonal basis which excludes the observed behaviouroriginating from the seasonal cycle

432 Chemical species within the advected aerosollayers

As a further step to adequately discriminate local and non-local sources we took advantage of the 1-year long (2017)chemical aerosol characterisation dataset in the urban site ofAostandashDowntown and applied the PMF to the three datasetsdescribed in Sect 33 (PMF datasets a b c ie anioncationonly anioncation with ECOC and anioncation with met-als respectively) Results of this analysis show the main fac-tors shaping the composition of PM10 at the investigated siteare those given in Fig 11 all details on this outcome beinggiven in Sect S4 The question addressed here is howeverhow aerosol transport from the Po Valley affects the chem-ical properties of PM10 sampled in Aosta In the prelimi-nary analysis of three case studies reported in the compan-ion paper we found that the advected layers were rich in ni-trates and sulfates (accounting together with ammonium for30 ndash40 of the total PM10 mass) This kind of secondaryinorganic aerosol was indeed found in high concentrationsin the Po basin by previous studies (eg Putaud et al 20022010 Carbone et al 2010 Larsen et al 2012 Saarikoskiet al 2012 Gilardoni et al 2014 Curci et al 2015) Herewe further tested on a longer dataset whether the sum ofthe nitrate- and sulfate-rich factors (ie secondary aerosols)could in fact represent a good chemical marker of aerosoltransport from the Po basin (eg Kukkonen et al 2008 Tanget al 2014) In this respect it is worth highlighting that thesePMF modes are not correlated with typical locally producedpollutants such as NOx and PAHs (this is revealed by the lowpercentage contribution of NOx and PAHs in nitrate-rich andsulfate-rich modes in Figs S3ndashS5) which therefore excludestheir local origin We then combined the chemical informa-tion (the sum of the secondary components) with the ALCclassification This was done for the year 2017 which is theonly overlapping period between anioncation analyses andALC profiles (see Table 1) Results are shown in Fig 12 Inparticular Fig 12a shows the time evolution of the mass con-centration carried by the secondary aerosol modes in whicheach date is coloured according to the six ALC classes iden-tified It can be noticed that almost all peak values occur dur-ing days of type E or F ie when the advected layer is morepersistent and able to strongly influence the local air qual-ity at the ground The very good correspondence betweenthe sum of the nitrate- and sulfate-rich mode contributionsand the ALC classification as well as the poor correlationbetween the latter and the role of the other PMF-identifiedmodes (not shown) suggests that the link between the sec-ondary components and the aerosol layer advection is notsimply due to particular weather conditions affecting both

Figure 9 Comparison between daily PM10 anomalies (Eq 5) reg-istered in AostandashDowntown and Donnas (40 km apart) relative to amonthly moving average Circle colour represents the ALC classifi-cation (see legend) and the ldquoXrdquo symbol represents the unclassifieddays The 1 1 and the best fit line are also represented on the plotPearsonrsquos correlation index between the two series is ρ = 073

Figure 10 (a) Daily PM10 cycle sampled by the OPC in AostandashSaint-Christophe during non-event days (class A) and days with ar-riving and leaving aerosol layers (classes C and D) plotted togetherwith the daily evolution of PAH as a proxy of the local sources(dashed line right vertical axis in ngmminus3) (b) Same as (a) usingthe TEOM in AostandashDowntown

variables Rather the coupling between the chemical and theALC information indicates that aerosols of secondary origindominate during the advection episodes from the Po basin

A summary of the contribution of secondary modes to thetotal PM10 as a function of the day type on a statistical basisis given in Fig 12b The MannndashWhitney test applied to thesedata confirms that the contribution of secondary pollutants issignificantly lower for class A compared to B (p = 000053times 10minus8) C compared to D (p = 6times 10minus8) D comparedto E (p = 9times 10minus7) and E compared to F (p = 6times 10minus7)Figure 12b also allows us to quantify the contribution of non-local sources using as a baseline the results obtained for classA (ie no advection shaded area in Fig 12a) Note that thisbaseline is very low on average about 3 microg mminus3 as expectedfrom the unfavourable conditions for secondary aerosol pro-duction in the area under investigation (eg low expectedemissions of ammonia frequent winds) and from the un-observed correlation between secondary aerosol components

and their gas-phase precursors The non-local contribution iscalculated as the average difference between the mass fromthe secondary modes and this baseline and amounts to about5 microg mminus3 on a yearly basis ie 24 of the total PM10 con-centration reconstructed from the PMF On a seasonal ba-sis the absolute impact of air mass transport depends on thecoupling between emissions (stronger in winter and weakerin summer) and weather regimes (eg thermal winds occur-ring more frequently in summerautumn with respect to win-terspring Sect 41) This results in a PM10 contribution ofnearly 6 microg mminus3 in winter and autumn 4 microg mminus3 in summerand 3 microg mminus3 in spring In terms of relative contribution thisalso depends on the ldquobackgroundrdquo PM10 levels in Aosta It istherefore highest in summer (32 ) when thermally drivenfluxes are more frequent and local emissions lower and low-est in winter (16 ) when thermal winds are less frequentand local emissions higher Intermediate relative values arefound in spring and summer (27 and 28 respectively) Inspecific episodes advections from the Po basin may still pro-duce an increase in PM10 concentrations of up to 50 microg mminus3ie exceeding alone the EU daily threshold The most im-portant compounds contributing to this increase are nitrateammonium and sulfate ie the species characterising thenitrate- and sulfate-rich PMF modes However since thesulfate-rich mode additionally includes organic matter (OMcf Figs S3ndashS5) the latter species is also relevant in the ad-vection episodes especially in summer when the nitrate-richmode has its lowest contribution This finding agrees withprevious works identifying the Po basin as a source of or-ganic aerosol (Putaud et al 2002 Matta et al 2003 Gilar-doni et al 2011 Perrone et al 2012 Saarikoski et al 2012Sandrini et al 2014 Bressi et al 2016 Khan et al 2016Costabile et al 2017)

The Po Valley origin of the secondary components wasalso further proved by coupling the chemical information toback-trajectories Figure 13a shows the result of the TSMusing the sum of the nitrate- and sulfate-rich modes as aconcentration variable at the receptor It clearly reveals thattrajectories crossing the Po basin have a much higher im-pact on secondary aerosol measured in Aosta compared to airparcels coming from the northern side of the Alps Interest-ingly this is not the case using different source factors fromPMF For example Fig 13b reports the same map but rela-tive to the traffic and heating mode which is clearly muchmore homogeneous compared to Fig 13a and essentiallyreflects the density distribution of the trajectories This be-haviour highlights the local origin of the PMF source factorsother than the two chosen as proxies of the advection (sec-ondary aerosols) As sensitivity tests similar maps withoutany weighting applied are reported in Fig S7 No substantialdifferences can be noticed

Figure 11 Percentage of aerosol mass concentration carried by each mode for the three PMF datasets considered in the analysis (PMFdataset a anioncation only PMF dataset b anioncation together with ECOC PMF dataset c anioncation together with metals) and theirrespective confidence intervals from the BS-DISP test Details are provided in Sect S4

Figure 12 (a) Temporal chart of the contribution of secondary (nitrate-rich and sulfate-rich) modes to the total PM10 The shaded arearepresents the estimated local production of secondary aerosol The difference between each dot and this baseline can thus be read as thenon-local contribution to the aerosol concentration in AostandashDowntown Since ion chemical speciation started in 2017 only 1 year of overlapwith the ALC is available at the moment (see Table 1) (b) Boxplot of the contribution of secondary (nitrate-rich and sulfate-rich) modes tothe total PM10 concentration as a function of the day type PMF dataset ldquoardquo was used for both plots

433 PMF of aerosol size distributions and links tochemical properties

Exploring the links between aerosol chemical and physicalproperties is useful to get insights into the atmospheric mech-anisms leading to particle formation and transformation dur-ing their transport to the Aosta Valley Therefore we in-vestigated correlations between the results of the chemical

analyses on samples collected in AostandashDowntown and par-ticle size distributions (PSDs) measured by the Fidas OPCin AostandashSaint-Christophe To ensure comparability betweenthese two datasets the particle volume distributions from theFidas OPC (normally extending up to sizes of 18 microm) wereweighted by a typical cut-off efficiency of PM10 samplinghead (Sect S6) We then performed a PMF analysis of thehourly averaged PSDs from the Fidas OPC (hereafter re-

Figure 13 (a) Output of the Concentration Field Trajectory Statistical Model using the sum of the contributions from PMF nitrate- andsulfate-rich modes as a concentration variable at the receptor (AostandashDowntown) (b) Concentration field based on the traffic and heatingmode from PMF Trajectories are cut at the borders of the COSMO-I2 domain

ferred to as size-PMF) Four principal modes were identi-fied Their relative contribution to the total particle volumeis shown in Fig 14a These modes are centred at 02 052 and 10 microm diameter respectively and will be thus referredto as 02 05 2 and 10 microm size-PMF modes in the follow-ing A similar figure showing their dependence on seasonwind speed and direction is also provided in Fig S10 Thenumber of factors (p) was chosen based on physical con-siderations if p gt 4 then the last mode (largest sizes) splitsinto sub-modes which are not relevant for the present studyconversely if fewer components (p lt 4) are retained theymerge into unphysical modes (eg very large and very fineparticles combined together)

The scores from the size-PMF were then averaged on adaily basis to be compared to the chemical PMF ones (here-after chem-PMF Sect S4) Figure 14 compares time seriesof selected chem-PMF (dataset a) and size-PMF results Itshows that

a the nitrate-rich mode from chem-PMF and the 05 micromsize-PMF mode correlate very strongly (ρ = 081Fig 14b)

b the sum of the sulfate-rich mode and traffic and heatingmode correlates very strongly with the 02 microm size-PMFmode (ρ = 082 Fig 14c) Note that the correlation in-dex decreases (ρ between 035 and 069) if the sulfate-rich mode and traffic and heating mode are comparedseparately with the 02 microm mode

c a moderate statistically significant correlation wasfound between the chem-PMF soil plus road saltingmodes and the size-PMF coarser modes (sum of 2 and10 microm modes Pearsonrsquos correlation index ρ = 059 p

valuelt 22times 10minus16 Fig 14d) This suggests that bothindicate the same source and likely non-exhaust traf-fic emissions such as abrasion from brakes tire wearand road (with typical sizes of 2ndash5 microm) and resuspen-sion from the surface (up to gt 10 microm) as also foundin other studies (eg Harrison et al 2012) This agree-ment between the two independent datasets is remark-able considering that the particles were sampled attwo different sites (AostandashSaint-Christophe and AostandashDowntown) with distinct environmental features andthat coarse particles usually travel for short ranges Itis also worth mentioning that the correlation betweensingle modes from chem-PMF (road salting or soil) andsize-PMF (2 or 10 microm) separately is weaker (ρ between026 and 055) than the one resulting from their sumFinally from a preliminary analysis based on back-trajectories and desert dust forecasts (NMMBBSC-Dust httpessbscesbsc-dust-daily-forecast last ac-cess 2 August 2019) the 2 microm component seems to beadditionally connected to deposition and possibly re-suspension of mineral Saharan dust in Aosta an effectalready observed in other urban areas in central Italy(Barnaba et al 2017)

A further interesting feature is the clear size separationof the 02 and 05 microm accumulation modes which likely re-sults from different aerosol formation mechanisms in theatmosphere condensationcoagulation from primary emis-sionsaging on the one hand (02 microm mode also known asldquocondensation moderdquo) and aqueous-phase processes (eg infog during the cold season) on the other hand (formingldquodroplet moderdquo aerosol particles of about 05 microm diametereg Seinfeld and Pandis 2006 Wang et al 2012 Costabile

Figure 14 (a) Modes identified by the PMF analysis applied on the PSDs measured by the Fidas OPC (size-PMF) (bndashd) Comparisonbetween the PMF scoresrsquo time series obtained from analysis of chemical speciation (chem-PMF on PMF dataset ldquoardquo showing mass left-hand vertical axes) and size distributions (size-PMF showing volume right-hand vertical axes) The three panels show (b) contributionsfrom chem-PMF nitrate-rich (black) and size-PMF 05 microm (green) modes (c) contributions from the sum of the chem-PMF sulfate-rich andtraffic and heating modes (black) and from the 02 microm size-PMF mode (blue) (d) sum of contributions from chem-PMF road salting and soilmodes (black) and from 2 and 10 microm size-PMF modes (red) Correlation values (ρ) are also reported in each plot

et al 2017) Finally the very good correlation between thefine particles and the nitrate- and sulfate-rich modes at thetwo stations is a further hint that these kinds of particles aremostly of non-local origin

Overall the good agreement between the size- and chem-PMF results further strengthens their independent outputs

44 Impact of Po Valley advections on northwesternAlps columnar aerosol properties

ALC measurements revealed the vertical extent and thick-ness of the advected polluted layers from the Po ValleyWe therefore also investigated whether and how much theselayers impact the column-integrated aerosol properties (iethose sounded by ground-based Sun photometers but alsoby satellites) To this purpose the ALC day-type classifica-tion was applied to the measurements of the AostandashSaint-Christophe Sun photometer The average AOD at 500 nmbinned by day type and season is shown in Fig 15a It canbe noticed that a general increase in the AOD is found fromclasses A (about 004 on average) to F for which AOD is

more than 4 times higher (018) The advections are alsofound to affect the Aringngstroumlm exponent (Eq 2 Fig 15b)representative of the mean particle size Results from thiscolumn-integrated perspective confirm that the transportedparticles are on average smaller than the locally producedones In fact the mean Aringngstroumlm exponents vary from 10for days A in winter and autumn up to 16 for days EndashFand show a general increase among the classes for every sea-son To confirm and fully understand this dependence of theAringngstroumlm exponents on the ALC classification we also in-vestigated the columnar volume PSDs retrieved from the Sunphotometer almucantar scans during the Po Valley pollutiontransport events This analysis (Fig S11) confirms what wealready found with the surface-level PSDs from the FidasOPC ie that the advections increase the number of parti-cles in all sizes notably in the sub-micron fraction This ex-plains the higher Aringngstroumlm exponents retrieved by the Sunphotometer during the advections

A further link between aerosol physical and chemicalproperties is explored through the variability of the sin-

Figure 15 (a) Average aerosol optical depth at 500 nm from thePOM Sun photometer for each day type (colour code as in Fig 12)and season (b) Aringngstroumlm exponent calculated in the 400ndash870 nmband (c) yearly averaged single scattering albedo at 500 nm B daysin spring were removed from panels (a) and (b) due to insufficientstatistics (only 3 d)

gle scattering albedo with day type (Fig 15c) Yearly aver-aged data are shown in this case due to the limited numberof data points available for this parameter (the almucantarplane must be free of clouds to accurately retrieve the SSASect 2) Figure 15c reveals that SSA at 500 nm generally in-creases from class A to class F (092 to 098) which agreeswith the fact that secondary and thus more scattering aerosolis transported with the advections This information is partic-ularly important for follow-up radiative transfer evaluationsunder the investigated conditions In this respect also notethat further analysis more focussed on the impact of theseadvections on particle absorption properties is planned forthe future

45 Comparison between observations and CTMsimulations

Accurate simulations of air-quality metrics are of utmost im-portance for environmental protection agencies Indeed notonly are they essential from a scientifictechnical perspective(contributing to better understanding the local and remotepollution dynamics) but they also represent a fundamental

duty towards the public air-quality forecasts being dissem-inated every day In this section we compare long-term (1-year 2017) daily averages of surface PM10 to the correspond-ing simulations by FARM in AostandashDowntown (similar re-sults are found for the other sites in the domain and are notreported here) and next to Ivrea This exercise was also aimedat evaluating whether and how the information gathered bythe ALC can be used to understand deficiencies and thus im-prove the model performances The 1-year time series of boththe measured and simulated PM10 in the AostandashDowntownand Ivrea sites are displayed in Fig 16 Model and mea-surement show similarities (such as the same seasonal cycleindicating that local emissions from the regional inventoryand general atmospheric processes are correctly reproduced)but also divergences (especially PM10 underestimation byFARM as already shown and discussed in the companionpaper) Table 3 summarises some statistical indicators of themodelndashmeasurement comparison namely mean bias error(MBE) root mean square error (RMSE) normalised meanbias error (NMBE) normalised mean standard deviation(NMSD ie (σMσOminus1) where σM and σO are the standarddeviations of the model and observation) and Pearsonrsquos cor-relation coefficient (ρ) The overall performances of FARMare comparable to the results obtained in other studies usingCTMs (eg Thunis et al 2012) For the AostandashDowntowncase we additionally explore the absolute (Fig 17a) and rel-ative (Fig 17b) differences between modelled and observedPM10 in relation to the ALC-derived classes These resultsreveal that the model underestimation mostly occurs in thecases of strongest advections (F) with extreme model devi-ations as low as minus40 microgmminus3 (Fig 17a) ie minus50 or lower(Fig 17b) The observed modelndashmeasurement discrepanciesmight originate from (1) an incomplete representation ofthe inventory sources (emission component) (2) inaccurateNWP modelling of the meteorological fields and notably thewind (transport component) or a combination of (1) and (2)Some details on these aspects are provided in the following

1 Emissions Inaccuracies in the (local and national)emission inventories could degrade the comparison be-tween simulations and observations Based on resultsshown in Fig 17a b we decided to investigate the sen-sitivity of our simulations in AostandashDowntown to themagnitude of the external contributions (boundary con-ditions) In particular we used a simplified approachto speed up the calculations assuming that the contri-bution from outside the regional domain and the localemissions add up without interacting we simulated thesurface PM10 concentrations turning onoff the bound-ary conditions (dark- and light-blue lines in Fig 16)to roughly estimate the only contribution from sourcesoutside the regional domain Interestingly the differ-ence between the two FARM runs correlates well withthe advection classes observed by the ALC (Fig 18)This clear correlation between the simulations and the

Figure 16 Long-term (1-year) comparison between PM10 surface measurements and simulations (FARM) in AostandashDowntown (a) andIvrea (b)

Table 3 Statistics of comparison between PM10 model forecasts and measurements at the site of AostandashDowntown Results with bothW = 1and W = 4 weighting factors of the PM fraction arriving from outside the boundaries of the regional domain are reported

W MBE (microgmminus3) RMSE (microgmminus3) NMBE () NMSD () ρ

1 minus7 15 minus35 minus16 0624 1 13 5 minus8 061

experimentally determined atmospheric conditions is afirst good indication that the NWP model used as in-put to the CTM yields reasonable meteorological in-puts to FARM Then to further explore the sensitivityto the boundary conditions we gradually increased bya weighting factorW the PM10 concentration from out-side the boundaries of the domain trying to match theobserved values In Fig 17c d we show the results ob-tained with W = 4 This exercise shows that the over-all mean bias error is much reduced compared to theoriginal simulations (W = 1 Fig 17a b) especially forthe winter and autumn seasons while slight overestima-tions are now visible for summer and spring Also theannually averaged MBE NMBE and NMSD improve(Table 3) whilst the other statistical indicators remainstable or even slightly worsen

Clearly our simplified test has major limitations Forexample seasonally and geographically dependent (ierelative to the position on the regional domain border)factors W should be used to better correct deficien-cies in the national inventory Also the high weight-ing factors needed to match the measurements in win-ter and autumn suggest not only underestimated emis-sions but also missing aerosol production mechanisms(eg aqueous-phase particle production as in fogcloudprocessing Gilardoni et al 2016) during the cold sea-son In fact this simplified exercise was only intended

to roughly estimate the magnitude of the ldquooutside PM10tuning factorrdquo necessary to match the observationsDespite this limitation this numerical experiment stillshows quite convincingly that biases in air-quality fore-casts can be reduced by revising the emission invento-ries on the basis of experimental data among them thosefrom unconventional air-quality systems (eg the ALCsused in this study)

2 Transport Although the COSMO-I2 grid spacing is cer-tainly in line with that of other state-of-the-art oper-ational limited-area models in our complex terrain itcould be insufficient to appropriately resolve local me-teorological phenomena triggered by the valley orog-raphy (Wagner et al 2014b) also considering that theactual model resolution is 6ndash8 times that of the grid cell(Skamarock 2004) For example Schmidli et al (2018)show that at a 22 km grid step the COSMO modelpoorly simulates valley winds while at a 11 km gridstep the diurnal cycle of the valley winds is well repre-sented Similarly Giovannini et al (2014) show that a2 km step can be considered the limit for a good repre-sentation of valley winds in narrow Alpine valleys Thesmoothed digital elevation model (DEM) used withinCOSMO and FARM could also play a direct role inthe detected underestimation In fact as mentioned inSect 32 the model surface altitude of the Aosta ur-

ban area is 900 m asl whereas the actual altitude isabout 580 m asl (Table 1) The adjacent cells are givenan even higher altitude owing to the fact that the val-ley floor and the neighbouring mountain slopes are notproperly resolved at the current resolution This resultsin an apparent lower elevation of the sites located in flat-ter and wider areas such as Aosta while the real to-pography profile at the bottom of the valley presents amonotonic increase from the Po plain to Aosta There-fore it is expected that just by better reproducing theorography a higher resolution would better resolve thehorizontal advection of aerosol-laden air from muchlower altitudes on the plain

Most of these issues were extensively addressed in thecompanion paper (Dieacutemoz et al 2019) In that studyCOSMO-I2 was shown to be capable of reproducing themountainndashplain wind patterns observed at the surfaceboth on average (cf Fig S1 in the companion paper)and in specific case studies (Fig S13 therein) Never-theless it was also found to slightly anticipate in timeand overestimate the easterly diurnal winds in the firsthours of the afternoon and to overestimate the nighttimedrainage winds (katabatic winds ventilating the urbanatmosphere and reducing pollutant loads) These limi-tations were mostly attributed to the finite resolution ofthe model

To disentangle the role of factors (1) and (2) discussedabove we evaluated the model performances in a flatter areaof the domain where simulations are expected to be less af-fected by the complex orography of the mountains We there-fore compared PM10 simulations and measurements in Ivrea(Fig 16b) This test is also useful for operating the CTMat the boundaries of the domain with special focus on theemissions from the ldquoboundary conditionsrdquo Model underes-timation in both the warm and cold seasons is evident In-cidentally it can be noticed that concentrations simulatedwithout ldquoboundary conditionsrdquo are nearly zero since the con-sidered cell is far from the strongest local sources and mostof the aerosol comes from outside the regional domain Asdone for Aosta (Fig 17a b) Fig 19a b present the ab-solute and relative differences between the model and theobservations in Ivrea The overall NMBE is minus073 whichrather well corresponds to the underestimation factor W = 4(or NMBE=minus075) found for AostandashDowntown and othersites in the valley Since it is expected that in this flat area theNWP model is able to better resolve the circulation comparedto the mountain valley this result suggests poor CTM perfor-mances to be mostly related to underestimated emissions atthe boundaries rather than to incorrect transport In additionto this general underestimation a large scatter between ob-servations and simulations can be noticed in Fig 16 the lin-ear correlation index between simulations and observation inIvrea being rather low (ρ = 054) If such a large scatter dueto the erroneous boundary conditions is already detectable

at the border of the domain it is likely that at least part of theRMSE reported in Table 3 for AostandashDowntown can be at-tributed to inaccuracies in the national inventory As alreadymentioned an additional contribution to the observed RMSEcould still be given by errors in modelling transport due tothe coarse resolution of the NWP model although we arenot able to quantify the relative role of the two factors Tounravel this issue high-resolution models (grid step 05 kmeg Golzio and Pelfini 2018 Golzio et al 2019) are beingtested on the investigated area and their results will be ad-dressed in future studies

Finally within the limits of the model performances dis-cussed above we use FARM to extend our observationalfindings to a wider domain compared to the few measuringsites In Fig 20 we provide some summarising plots of 1-year simulations In particular these show the 3-D simulatedfields of PM10 concentrations in terms of both surface-level(a b and c) and vertical transects along the main valley (de and f) averaged over all non-advection and advection daysin 2017 In this latter case simulations with W = 1 (b e)andW = 4 (c f) are included Compared to the ldquobackgroundconditionrdquo (ALC type A) the advection cases show higherPM10 concentrations and a different geographical distribu-tion increasing towards the entrance of the valley (Fig 20be) Obviously the absolute PM10 loads are higher when thenon-local contribution is multiplied by W = 4 (Fig 20c f)which as discussed above is the W value providing a bettermatching with the PM10 concentrations actually measured inAosta and Ivrea Vertical profiles of simulated concentrationsare also evidently impacted by the advections (eg Fig 20dndashf) A more complete set of maps is provided in Fig S12in which the contribution of particulate produced insidethe regional domain (local sources) and outside the domain(boundary conditions) has been partitioned An interestingfeature in Fig 20 is that the average maps of local PM10do not change between advection or non-advection caseswhile the (relative) concentration maps of aerosol comingfrom outside the domain show noticeable differences thusconfirming once again that the polluted air masses detectedby the ALC in AostandashSaint-Christophe are of non-local ori-gin In conclusion the excellent correspondence between themodel output and the experimental classification based onthe ALC ie the remarkable difference of the concentrationsand their vertical distribution between days of advection andnon-advection highlights both the rather good performancesof the CTM and the ability of the measurement dataset usedto reveal the phenomenon

5 Conclusions and perspectives

In this work we used long-term (2015ndash2017) multi-sensorobservational datasets (Table 1) coupled to modelling toolsto describe pollution aerosol transport from the Po basin tothe northwestern Alps Key to this study was the deployment

Figure 17 Absolute (a c) and relative (b d) differences between simulated and observed PM10 concentrations at the surface in AostandashDowntown The mean bias error (MBE) and the normalised mean bias error (NMBE) for each case are reported in the plot titles (ab) FARM simulations as currently performed by ARPA (c d) The PM10 concentrations from outside the boundaries of the domain weremultiplied by a factor W = 4

in Aosta of an automated lidar ceilometer (ALC) with its op-erational (247) capabilities This allowed us to (a) collectand present the first long-term dataset of aerosol vertical dis-tribution in the northwestern Alps (b) provide observationalevidence of the complex structure and dynamics of the at-mosphere in the Alpine valleys and (c) stimulate the investi-gation of the phenomenon on a 4-D scale with complement-ing observational and modelling tools The analysis over thelong term allowed us to expand the investigation of specificcase studies provided in the companion paper (Dieacutemoz et al2019) and answer some key scientific questions (as listed inthe Introduction)

1 Driving meteorological factors and frequency of theaerosol pollution transport events The newly developed

classification based on the ALC profiles (928 classifieddays Fig 3) permitted us to aggregate the days withsimilar aerosol vertical structures and examine the aver-age properties of each group Notably we could under-stand the link between the wind flow regimes and theaerosol transport The frequency of the elevated layerswas observed to be on average 43 ndash50 of the daysdepending on the season (ie slightly less than 60 ofthe days falling in an ALC class different from ldquoOtherrdquoin winter and spring to more than 70 in summer)and is clearly connected to eastern wind flows suchas plain-to-mountain thermally driven winds (blowingnearly 70 of the days in summer Fig 4) This waseven clearer using trajectory statistical models (Fig 13)

Figure 18 Difference between PM10 surface concentrations inAostandashDowntown simulated by FARM taking the boundary condi-tions into account (FARMtot) and without taking them into account(FARMlocal only local sources) as a function of the advection classobserved by the ALC

and contingency tables (Fig 5) showing the clear con-nections between the arrival of an elevated aerosol layerover the northwestern Alps and the wind direction Suchlayers were observed 93 of the days with easterlywinds (Fig 5a) with increasing probability of occur-rence for increasing air mass residence time over the Poplain (aerosol layer detected on over 80 of days whenthe trajectory residence times in the Po basin exceed 1 hand up to 100 of days with air mass residence timegt 10 h Fig 5b)

2 Advected aerosol physical and chemical properties Thegeneral spatio-temporal characteristics of the aerosollayers were statistically assessed based on the multi-year ALC series The top altitudes of the layer rangefrom 500 m to nearly 4000 m agl depending on seasonand according to the convective boundary layer heightNotably the high altitudes reached by the aerosol layerin summer are compatible with transport and deposi-tion of other contaminants such as pesticides (Rizziet al 2019) and micro-plastics (Allen et al 2019 Am-brosini et al 2019) on glaciers snow fields and remotemountain catchments Also a dataset of aerosol layerheights may be valuable for follow-up studies on the ra-diative forcing of particulate matter in connection withthe elevation-dependent warming in some mountain re-gions (Pepin et al 2015)

The duration of the phenomenon ranges from a fewhours to several days in cases of atmospheric stabilityThe mean optical and microphysical properties of theaerosol layers were further sounded using a Sun pho-tometer This showed that the columnar aerosol prop-erties are all impacted by the pollution transport AOD

at 500 nm increases from 004 on non-event days up to018 on event days on average relevant values of theAringngstroumlm exponent (400ndash870 nm) and single scatteringalbedo (SSA at 500 nm) change from 10 to 16 andfrom 092 to 098 respectively and depend on the du-rationseverity of the advection (Fig 15) This indicatesthat the transported particles are on average smaller andmore light-scattering than the locally produced ones andagrees well with the results of measurements and anal-yses of aerosol properties at the surface Positive matrixfactorisation (PMF) of chemical properties at surfacelevel revealed that particles advected from the Po Valleyto the northwestern Alps are mostly of secondary origin(eg Fig 12) and mainly composed of nitrates sulfatesand ammonium Interestingly we also found an organiccomponent in the warm season which confirms the PoValley as an OM source Moreover particle size distri-butions showed increase in the fine fraction (lt 1 microm)during Po Valley advection days Correlation of chem-ical PMF with independent PMF performed on aerosolsize distribution also provided evidence of a clear linkbetween chemical properties and aerosol size (correla-tion indices ρ gt 08 Fig 14) which proves that differ-ent aerosol formation mechanisms act in the atmospherein different seasons In particular we found some evi-dence of aqueous processing of particles in winter (egfog andor cloud processing) through identification of aPMF ldquodroplet moderdquo in the winter results

3 Aerosol transport impact on air quality At the bottomof the valley the advections were demonstrated to al-ter both the absolute concentrations of PM10 (these be-ing up to 35 times higher during event days comparedto non-event days Fig 8) and their daily cycles withhourly variations of up to 30 microgmminus3 (Fig 10) PMFstatistics on chemical data identified five to seven differ-ent sources depending on the available chemical char-acterisation two of which (nitrate-rich mode in winterand sulfate-rich mode in summer) were attributed to thearrival of particles from outside the Aosta Valley as alsoconfirmed by trajectory statistical models (Fig 13) Us-ing their relevant scores as a proxy we estimate an aver-age 25 contribution of these transported nitrate- andsulfate-rich components to the Aosta PM10 This impactvaries on a seasonal basis The relative contribution ofnon-local PM10 is highest in summer (32 ) when ad-vection is most frequent and local PM10 is lowest whileit is lowest in winter (16 ) when advection is least fre-quent and local PM10 is highest In absolute terms thereverse occurs and the impact of transport is found to behighest in winterautumn reaching levels of 50 microgmminus3

in some episodes (thus exceeding alone the EU legisla-tive limits) This occurs due to the superposition of ad-vected particles with higher background concentrationsin the coldest part of the year when emissions are high-

Figure 19 Absolute (a) and relative (b) differences between simulated and observed PM10 concentrations at the surface in Ivrea

Figure 20 Average 3-D fields of PM10 concentration simulated by FARM extracted at the surface level (a b c) and on a vertical transect (de f) (a) Average of all non-advection days (categorised as ldquoArdquo by the experimental ALC classification) (b) Average of advection days (ldquoCrdquoldquoErdquo and ldquoFrdquo from the ALC classification) using a weighting factor W = 1 for the PM10 contribution coming from outside the boundariesof the regional domain (c) Same as (b) using W = 4 as a weighting factor (d) (e) and (f) are vertical transects along the main valley (iealong the dotted line in a to c) corresponding to the three cases above The altitude of the three measuring sites shown in the panels is theone from the digital elevation model which is a smoothed representation of the real topography The white area in the second row of panelsrepresents the surface The advections investigated in this study come from the east (right side of all the panels)

est and pollution is further enhanced by adverse weatherconditions ie low-level inversions locally worsenedby the valley topography In this study the impact ofPo Valley advection on air quality was mostly quanti-fied for the AostandashDowntown station In rural and re-mote sites of the region the non-local contribution is ex-pected to be even larger in relative terms Increasingthe number of stations collecting samples for chemicalanalyses would be an important follow-up to estimatethe non-local contribution to PM10 in non-urban sitesThe good correlation found between the PM10 concen-tration anomalies in the urban site of AostandashDowntownand in the regional site of Donnas (ρ gt 07 Fig 9) ishowever a clear indication that aerosol loads all over theregion are mainly influenced by trans-regional transportrather than by local emissions It is also worth mention-ing that the aerosol particle number (here only measuredin the 018ndash18 microm range ie missing the finest particles)increased remarkably (up to gt 3000 particles cmminus3) insome transport episodes with possible implications forhuman health Some preliminary results (Sect S5) alsoindicate that this regular air mass transport is also able toalter the oxidative capacity of the atmosphere (NOxNOratio) although further investigation is needed to betterquantify it

In general for both air-quality and climate evaluationsthe identification of monitoring sites (and time periods)representative of ldquobackground conditionsrdquo is a criticalissue to differentiate local and regional pollution lev-els (eg Yuan et al 2018) A global network of atmo-spheric ldquobaselinerdquo monitoring stations is for examplemaintained within the World Meteorological Organisa-tionrsquos (WMO) Global Atmosphere Watch (GAW) pro-gramme (WMO 2017) with the intention of provid-ing regionally representative measurements relativelyfree of significant local pollution sources Our observed(PM10) daily cycle (Figs 2 6 and 10) indicates thatcaution should be paid when assuming mountain sitesto be representative of background conditions In par-ticular we show that the influence of nearby pollutionsources in high-altitude sites might not be limited to thecentral hours of the day only (when convection-drivengrowth of the nearby plain mixing layer is known to up-lift pollutants) but rather extend to night-time measure-ments via the mountainndashvalley circulation and particlegrowth mechanisms addressed in this study

4 Ability to predict the arrival and impacts of polluted airmass transport Experimental data and their couplingwith modelling tools well demonstrated the Po plainorigin of the polluted layers and their increased prob-ability of detection as a function of the air massesrsquo res-idence time over the origin region Current chemicaltransport models however were found to reproduce thephenomenon in qualitative terms but manifest both sys-

tematic underestimations of the absolute advected PM10(biases) and large scatter to the measurements Based ontwo 1-year long simulated datasets in AostandashDowntownand Ivrea we tested how the air-quality forecasts couldbe improved by updating the emission inventories andusing the ALC to constrain the modelled emissions Af-ter some sensitivity tests we tuned the national inven-tory to better match the measurements (ldquoboundary con-ditionsrdquo increased by a factor of 4) In this case themodel underestimation was reduced from biasesltminus40tominus20 microgmminus3 in the worst cases with mean average bi-ases lt 2 microgmminus3 (Table 3) thus enhancing on averageour ability to correctly predict the PM concentrationand their relevant impacts (Fig 20) Still further im-provements are necessary to enhance the modelrsquos abil-ity to better reproduce aerosol processes (eg aqueous-phase chemistry Gong et al 2011) and wind fields us-ing higher-resolution NWP models

In conclusion we demonstrated that despite being consid-ered a pristine environment the northwestern Alps are regu-larly affected by a sort of ldquopolluted aerosol tiderdquo ie by awind-driven transport of polluted aerosol layers from the Poplain Overall this calls for air pollution mitigation policiesacting at least over the regional scale in this fragile environ-ment

Data availability The ALC data are available upon re-quest from the Alice-net (alicenetisaccnrit) and E-PROFILE (httpdatacedaacukbadceprofiledata E-PROFILE 2019) networks The Sun photometer data canbe downloaded from the EuroSkyRad network web site(httpwwweuroskyradnetindexhtml EuroSkyRad 2019)after authentication (credentials may be requested to M Campan-elli mcampanelliisaccnrit) The measurements from the ARPAValle drsquoAosta air-quality surface network are available on theweb page httpwwwarpavdaititariala-qualitagrave-dell-ariastazioni-di-monitoraggioinquinanti-export-dati (ARPAValle drsquoAosta 2019) The PM10 measurements in Ivreaare available on the Piedmont regional governmentweb page httpwwwregionepiemonteitambienteariarilevariadayariaweb-newindexphpithome (RegionePiemonte 2019) The weather data from the AostandashSaint-Christophe and Saint-Denis stations can be retrieved fromhttpcfregionevdaitrichiesta_datiphp (Centro Funzionale2019) upon request to Centro Funzionale della Valle drsquoAosta Therest of the data can be requested from the corresponding author(hdiemozarpavdait)

Supplement The supplement related to this article is available on-line at httpsdoiorg105194acp-19-10129-2019-supplement

Author contributions HD FB and GPG conceived and designedthe study interpreted the results and wrote the paper HD analysed

the data TM supplied the meteorological observations and numeri-cal weather predictions GP performed the chemical transport sim-ulations SP carried out the ECOC analyses and IKFT helped withthe interpretation of the chemical speciation MC supplied the POMcalibration factors

Competing interests The authors declare that they have no conflictof interest

Special issue statement SKYNET ndash the international network foraerosol clouds and solar radiation studies and their applica-tions (AMTACP inter-journal SI) SI statement this article is partof the special issue ldquoSKYNET ndash the international network foraerosol clouds and solar radiation studies and their applications(AMTACP inter-journal SI)rdquo It is not associated with a confer-ence

Acknowledgements The authors would like to thank Alessan-dra Brunier Giuliana Lupato Paolo Proment Stefania VaccariMaria Cristina Gibellino (ARPA Valle drsquoAosta) for carrying outthe chemical analyses Marco Pignet Claudia Tarricone andManuela Zublena (ARPA Valle drsquoAosta) for providing the data fromthe air-quality surface network and Paolo Lazzeri (APPA Trento)for helping with the PMF analysis The authors would like to ac-knowledge the valuable contribution of the discussions in the work-ing group meetings organised by COST Action ES1303 (TOPROF)They also gratefully acknowledge the Institute for Atmospheric andClimate Science ETH Zurich Switzerland for the provision of theLAGRANTO software used in this publication ARPA Piemonteand Regione Piemonte for the PM10 measurements at the Ivrea sta-tion and two anonymous reviewers whose comments helped im-prove and clarify the manuscript

Review statement This paper was edited by Pedro Jimenez-Guerrero and reviewed by two anonymous referees

References

Agnesod G Moulin P-A Pession G Villard H andZublena M Eacutetude Air Espace Mont-Blanc ndash RapportTechnique Tech rep Espace Mont Blanc available athttpwwwarpavdaitimagesstoriesARPAariaprogettiprogairmb_agnesod_2003pdf (last access 2 August 2019)2003

Allen S Allen D Phoenix V R Le Roux G Duraacuten-tez Jimeacutenez P Simonneau A Binet S and Galop DAtmospheric transport and deposition of microplastics ina remote mountain catchment Nat Geosci 12 339ndash344httpsdoiorg101038s41561-019-0335-5 2019

Aringngstroumlm A On the Atmospheric Transmission of Sun Radiationand on Dust in the Air Geogr Ann 11 156ndash166 1929

Ambrosini R Azzoni R S Pittino F Diolaiuti G Franzetti Aand Parolini M First evidence of microplastic contamination in

the supraglacial debris of an alpine glacier Environ Pollut 253297ndash301 httpsdoiorg101016jenvpol201907005 2019

ARPA Valle drsquoAosta ARPA Valle drsquoAosta Air quality dataavailable at httpwwwarpavdaititariala-qualitagrave-dell-ariastazioni-di-monitoraggioinquinanti-export-dati last access 2August 2019

Arvani B Bradley Pierce R Lyapustin A I Wang Y Gher-mandi G and Teggi S Seasonal monitoring and estimation ofregional aerosol distribution over Po valley northern Italy usinga high-resolution MAIAC product Atmos Environ 141 106ndash121 httpsdoiorg101016jatmosenv201606037 2016

Ashbaugh L L Malm W C and Sadeh W Z A resi-dence time probability analysis of sulfur concentrations atgrand Canyon National Park Atmos Environ 19 1263ndash1270httpsdoiorg1010160004-6981(85)90256-2 1985

Baldauf M Seifert A Foumlrstner J Majewski D Raschendor-fer M and Reinhardt T Operational Convective-Scale Nu-merical Weather Prediction with the COSMO Model Descrip-tion and Sensitivities Mon Weather Rev 139 3887ndash3905httpsdoiorg101175MWR-D-10-050131 2011

Barnaba F and Gobbi G P Aerosol seasonal variability overthe Mediterranean region and relative impact of maritime conti-nental and Saharan dust particles over the basin from MODISdata in the year 2001 Atmos Chem Phys 4 2367ndash2391httpsdoiorg105194acp-4-2367-2004 2004

Barnaba F Gobbi G P and de Leeuw G Aerosol strat-ification optical properties and radiative forcing in Venice(Italy) during ADRIEX Q J Roy Meteor Soc 133 47ndash60httpsdoiorg101002qj91 2007

Barnaba F Putaud J P Gruening C dellrsquoAcqua A andDos Santos S Annual cycle in co-located in situ total-columnand height-resolved aerosol observations in the Po Valley (Italy)Implications for ground-level particulate matter mass concen-tration estimation from remote sensing J Geophys Res 115D19209 httpsdoiorg1010292009JD013002 2010

Barnaba F Bolignano A Di Liberto L Morelli M Lu-carelli F Nava S Perrino C Canepari S Basart SCostabile F Dionisi D Ciampichetti S Sozzi R andGobbi G P Desert dust contribution to PM10 loads inItaly Methods and recommendations addressing the rele-vant European Commission Guidelines in support to the AirQuality Directive 200850 Atmos Environ 161 288ndash305httpsdoiorg101016jatmosenv201704038 2017

Belis C Blond N Bouland C Carnevale C Clappier ADouros J Fragkou E Guariso G Miranda A I NahorskiZ Pisoni E Ponche J-L Thunis P Viaene P and Volta MStrengths and Weaknesses of the Current EU Situation SpringerInternational Publishing 69ndash83 httpsdoiorg101007978-3-319-33349-6_4 2017

Bigi A and Ghermandi G Long-term trend and variabilityof atmospheric PM10 concentration in the Po Valley AtmosChem Phys 14 4895ndash4907 httpsdoiorg105194acp-14-4895-2014 2014

Bigi A and Ghermandi G Trends and variability of atmosphericPM25 and PM10ndash25 concentration in the Po Valley Italy AtmosChem Phys 16 15777ndash15788 httpsdoiorg105194acp-16-15777-2016 2016

Binkowski F Science Algorithms of the EPA Models-3 Com-munity Multiscale Air Quality (CMAQ) Modeling System vol

EPA600R-99030 in The aerosol portion of Models-3 CMAQedited by Byun D W and Ching J K S 1999

Birch M E and Cary R A Elemental Carbon-BasedMethod for Monitoring Occupational Exposures to Partic-ulate Diesel Exhaust Aerosol Sci Tech 25 221ndash241httpsdoiorg10108002786829608965393 1996

Blyth S Mountain watch environmental change amp sustainabledevelopmental in mountains United Nations Environment Pro-gramme 2002

Bonvalot L Tuna T Fagault Y Jaffrezo J-L Jacob VChevrier F and Bard E Estimating contributions frombiomass burning fossil fuel combustion and biogenic carbonto carbonaceous aerosols in the Valley of Chamonix a dual ap-proach based on radiocarbon and levoglucosan Atmos ChemPhys 16 13753ndash13772 httpsdoiorg105194acp-16-13753-2016 2016

Bourgeois I Savarino J Caillon N Angot H BarberoA Delbart F Voisin D and Cleacutement J-C Tracingthe Fate of Atmospheric Nitrate in a Subalpine Water-shed Using 117O Environ Sci Technol 52 5561ndash5570httpsdoiorg101021acsest7b02395 2018

Bressi M Sciare J Ghersi V Mihalopoulos N Petit J-ENicolas J B Moukhtar S Rosso A Feacuteron A Bonnaire NPoulakis E and Theodosi C Sources and geographical ori-gins of fine aerosols in Paris (France) Atmos Chem Phys 148813ndash8839 httpsdoiorg105194acp-14-8813-2014 2014

Bressi M Cavalli F Belis C A Putaud J-P Froumlhlich RMartins dos Santos S Petralia E Preacutevocirct A S H BericoM Malaguti A and Canonaco F Variations in the chem-ical composition of the submicron aerosol and in the sourcesof the organic fraction at a regional background site of thePo Valley (Italy) Atmos Chem Phys 16 12875ndash12896httpsdoiorg105194acp-16-12875-2016 2016

Brulfert G Chemel C Chaxel E Chollet J-P JouveB and Villard H Assessment of 2010 air quality intwo Alpine valleys from modelling Weather type andemission scenarios Atmos Environ 40 7893ndash7907httpsdoiorg101016jatmosenv200607021 2006

Bucci S Cristofanelli P Decesari S Marinoni A SandriniS Groumlszlig J Wiedensohler A Di Marco C F NemitzE Cairo F Di Liberto L and Fierli F Vertical distribu-tion of aerosol optical properties in the Po Valley during the2012 summer campaigns Atmos Chem Phys 18 5371ndash5389httpsdoiorg105194acp-18-5371-2018 2018

Burkhardt J Zinsmeister D Grantz D A Vidic S SuttonM A Hunsche M and Pariyar S Camouflaged as degradedwax hygroscopic aerosols contribute to leaf desiccation treemortality and forest decline Environ Res Lett 13 085001httpsdoiorg1010881748-9326aad346 2018

Centro Funzionale Centro Funzionale weather data availableat httpcfregionevdaitrichiesta_datiphp last access 2 Au-gust 2019

Calori G Silibello C and Marras G FARM (Flexible Air qual-ity Regional Model) Model formulation and userrsquos Manual Ari-anet 47 edn 2014

Campana M Li Y Staehelin J Prevot A S Bona-soni P Loetscher H and Peter T The influence ofsouth foehn on the ozone mixing ratios at the high

alpine site Arosa Atmos Environ 39 2945ndash2955httpsdoiorg101016jatmosenv200501037 2005

Campanelli M Estelleacutes V Tomasi C Nakajima T MalvestutoV and Martiacutenez-Lozano J A Application of the SKYRADImproved Langley plot method for the in situ calibrationof CIMEL Sun-sky photometers Appl Opt 46 2688ndash2702httpsdoiorg101364AO46002688 2007

Campanelli M Mascitelli A Sanograve P Dieacutemoz H EstelleacutesV Federico S Iannarelli A M Fratarcangeli F Maz-zoni A Realini E Crespi M Bock O Martiacutenez-LozanoJ A and Dietrich S Precipitable water vapour contentfrom ESRSKYNET sun-sky radiometers validation againstGNSSGPS and AERONET over three different sites in EuropeAtmos Meas Tech 11 81ndash94 httpsdoiorg105194amt-11-81-2018 2018

Carbone C Decesari S Mircea M Giulianelli L FinessiE Rinaldi M Fuzzi S Marinoni A Duchi R PerrinoC Sargolini T Vardegrave M Sprovieri F Gobbi G An-gelini F and Facchini M Size-resolved aerosol chemicalcomposition over the Italian Peninsula during typical sum-mer and winter conditions Atmos Environ 44 5269ndash5278httpsdoiorg101016jatmosenv201008008 2010

Carslaw K S Boucher O Spracklen D V Mann G W RaeJ G L Woodward S and Kulmala M A review of naturalaerosol interactions and feedbacks within the Earth system At-mos Chem Phys 10 1701ndash1737 httpsdoiorg105194acp-10-1701-2010 2010

Cavalli F Viana M Yttri K E Genberg J and Putaud J-PToward a standardised thermal-optical protocol for measuring at-mospheric organic and elemental carbon the EUSAAR protocolAtmos Meas Tech 3 79ndash89 httpsdoiorg105194amt-3-79-2010 2010

Cesaroni G Badaloni C Gariazzo C Stafoggia M SozziR Davoli M and Forastiere F Long-term exposure to ur-ban air pollution and mortality in a cohort of more than a mil-lion adults in Rome Environ Health Persp 121 324ndash331httpsdoiorg101289ehp1205862 2013

Charron A Harrison R M Moorcroft S and BookerJ Quantitative interpretation of divergence between PM10and PM25 mass measurement by TEOM and gravimet-ric (Partisol) instruments Atmos Environ 38 415ndash423httpsdoiorg101016jatmosenv200309072 2004

Chazette P Couvert P Randriamiarisoa H Sanak J Bon-sang B Moral P Berthier S Salanave S and Tous-saint F Three-dimensional survey of pollution during win-ter in French Alps valleys Atmos Environ 39 1035ndash1047httpsdoiorg101016jatmosenv200410014 2005

Chemel C Arduini G Staquet C Largeron Y LegainD Tzanos D and Paci A Valley heat deficit as abulk measure of wintertime particulate air pollution inthe Arve River Valley Atmos Environ 128 208ndash215httpsdoiorg101016jatmosenv201512058 2016

Chu D A Kaufman Y J Zibordi G Chern J D Mao J LiC and Holben B N Global monitoring of air pollution overland from the Earth Observing System-Terra Moderate Reso-lution Imaging Spectroradiometer (MODIS) J Geophys Res108 4661 httpsdoiorg1010292002JD003179 2003

Clerici M and Meacutelin F Aerosol direct radiative effect in the PoValley region derived from AERONET measurements Atmos

Chem Phys 8 4925ndash4946 httpsdoiorg105194acp-8-4925-2008 2008

Cong Z Kawamura K Kang S and Fu P Penetration ofbiomass-burning emissions from South Asia through the Hi-malayas new insights from atmospheric organic acids Sci Rep5 9580 httpsdoiorg101038srep09580 2015

Costabile F Gilardoni S Barnaba F Di Ianni A Di Liberto LDionisi D Manigrasso M Paglione M Poluzzi V RinaldiM Facchini M C and Gobbi G P Characteristics of browncarbon in the urban Po Valley atmosphere Atmos Chem Phys17 313ndash326 httpsdoiorg105194acp-17-313-2017 2017

Cugerone K De Michele C Ghezzi A Gianelle V andGilardoni S On the functional form of particle numbersize distributions influence of particle source and mete-orological variables Atmos Chem Phys 18 4831ndash4842httpsdoiorg105194acp-18-4831-2018 2018

Curci G Ferrero L Tuccella P Barnaba F Angelini F Bolza-cchini E Carbone C Denier van der Gon H A C Fac-chini M C Gobbi G P Kuenen J P P Landi T C Per-rino C Perrone M G Sangiorgi G and Stocchi P Howmuch is particulate matter near the ground influenced by upper-level processes within and above the PBL A summertime casestudy in Milan (Italy) evidences the distinctive role of nitrate At-mos Chem Phys 15 2629ndash2649 httpsdoiorg105194acp-15-2629-2015 2015

Decesari S Allan J Plass-Duelmer C Williams B J PaglioneM Facchini M C OrsquoDowd C Harrison R M Gietl J KCoe H Giulianelli L Gobbi G P Lanconelli C CarboneC Worsnop D Lambe A T Ahern A T Moretti F Tagli-avini E Elste T Gilge S Zhang Y and DallrsquoOsto M Mea-surements of the aerosol chemical composition and mixing statein the Po Valley using multiple spectroscopic techniques AtmosChem Phys 14 12109ndash12132 httpsdoiorg105194acp-14-12109-2014 2014

de Freitas C R Tourism climatology evaluating environmentalinformation for decision making and business planning in therecreation and tourism sector Int J Biometeorol 48 45ndash54httpsdoiorg101007s00484-003-0177-z 2003

De Wekker S F J and Kossmann M ConvectiveBoundary Layer Heights Over Mountainous TerrainndashA Review of Concepts Front Earth Sci 3 77httpsdoiorg103389feart201500077 2015

Dhungel S Kathayat B Mahata K and Panday A Trans-port of regional pollutants through a remote trans-Himalayanvalley in Nepal Atmos Chem Phys 18 1203ndash1216httpsdoiorg105194acp-18-1203-2018 2018

Dieacutemoz H Siani A M Casale G R di Sarra A Serpillo BPetkov B Scaglione S Bonino A Facta S Fedele F Gri-foni D Verdi L and Zipoli G First national intercomparisonof solar ultraviolet radiometers in Italy Atmos Meas Tech 41689ndash1703 httpsdoiorg105194amt-4-1689-2011 2011

Dieacutemoz H Campanelli M and Estelleacutes V One Year ofMeasurements with a POM-02 Sky Radiometer at an AlpineEuroSkyRad Station J Meteorol Soc Jpn 92A 1ndash16httpsdoiorg102151jmsj2014-A01 2014a

Dieacutemoz H Siani A M Redondas A Savastiouk V McElroyC T Navarro-Comas M and Hase F Improved retrieval ofnitrogen dioxide (NO2) column densities by means of MKIV

Brewer spectrophotometers Atmos Meas Tech 7 4009ndash4022httpsdoiorg105194amt-7-4009-2014 2014b

Dieacutemoz H Barnaba F Magri T Pession G Dionisi D Pit-tavino S Tombolato I K F Campanelli M Della Ceca L SHervo M Di Liberto L Ferrero L and Gobbi G P Trans-port of Po Valley aerosol pollution to the northwestern Alps ndashPart 1 Phenomenology Atmos Chem Phys 19 3065ndash3095httpsdoiorg105194acp-19-3065-2019 2019

Dionisi D Barnaba F Dieacutemoz H Di Liberto L and Gobbi GP A multiwavelength numerical model in support of quantitativeretrievals of aerosol properties from automated lidar ceilome-ters and test applications for AOT and PM10 estimation At-mos Meas Tech 11 6013ndash6042 httpsdoiorg105194amt-11-6013-2018 2018

Dosio A Galmarini S and Graziani G Simulation of the cir-culation and related photochemical ozone dispersion in the Poplains (northern Italy) Comparison with the observations of ameasuring campaign J Geophys Res 107 LOP 2-1ndashLOP 2-24 httpsdoiorg1010292000JD000046 2002

EEA Air Quality in Europe ndash 2015 Report Tech rephttpsdoiorg10280062459 2015

E-PROFILE ALC data available at httpdatacedaacukbadceprofiledata last access 2 August 2019

EuroSkyRad Sun-sky radiometer data available at httpwwweuroskyradnetindexhtml last access 2 August 2019

Egger J Bajrachaya S Egger U Heinrich R Reuder JShayka P Wendt H and Wirth V Diurnal Winds in theHimalayan Kali Gandaki Valley Part I Observations MonWeather Rev 128 1106ndash1122 httpsdoiorg1011751520-0493(2000)128lt1106DWITHKgt20CO2 2000

EMEP Transboundary particulate matter photo-oxidants acidify-ing and eutrophying components Tech rep MSC-W CCC andCEIP 2016

EU Commission Directive 200850EC of the European Parliamentand of the Council of 21 May 2008 on ambient air quality andcleaner air for Europe Official Journal of the European UnionL1521ndash44 2008

EU Commission European Commission ndash Press release Air qual-ity Commission takes action to protect citizens from air pollu-tion Brussels 17 May 2018 Press Release Database availableat httpeuropaeurapidpress-release_IP-18-3450_enhtm (lastaccess 2 August 2019) 2018

Fernald F G Analysis of atmospheric lidar obser-vations some comments Appl Opt 23 652ndash653httpsdoiorg101364AO23000652 1984

Ferrero L Perrone M G Petraccone S Sangiorgi G FerriniB S Lo Porto C Lazzati Z Cocchi D Bruno F GrecoF Riccio A and Bolzacchini E Vertically-resolved particlesize distribution within and above the mixing layer over theMilan metropolitan area Atmos Chem Phys 10 3915ndash3932httpsdoiorg105194acp-10-3915-2010 2010

Ferrero L Castelli M Ferrini B S Moscatelli M Perrone MG Sangiorgi G DrsquoAngelo L Rovelli G Moroni B Scar-dazza F Mocnik G Bolzacchini E Petitta M and Cappel-letti D Impact of black carbon aerosol over Italian basin val-leys high-resolution measurements along vertical profiles ra-

diative forcing and heating rate Atmos Chem Phys 14 9641ndash9664 httpsdoiorg105194acp-14-9641-2014 2014

Finardi S Silibello C DrsquoAllura A and Radice P Anal-ysis of pollutants exchange between the Po Valley and thesurrounding European region Urban Clim 10 682ndash702httpsdoiorg101016juclim201402002 2014

Fuzzi S Baltensperger U Carslaw K Decesari S Denier vander Gon H Facchini M C Fowler D Koren I LangfordB Lohmann U Nemitz E Pandis S Riipinen I RudichY Schaap M Slowik J G Spracklen D V Vignati EWild M Williams M and Gilardoni S Particulate matterair quality and climate lessons learned and future needs At-mos Chem Phys 15 8217ndash8299 httpsdoiorg105194acp-15-8217-2015 2015

Gariazzo C Silibello C Finardi S Radice P Piersanti ACalori G Cecinato A Perrino C Nussio F Cagnoli MPelliccioni A Gobbi G P and Di Filippo P A gasaerosol airpollutants study over the urban area of Rome using a comprehen-sive chemical transport model Atmos Environ 41 7286ndash7303httpsdoiorg101016jatmosenv200705018 2007

Gilardoni S Vignati E Cavalli F Putaud J P Larsen BR Karl M Stenstroumlm K Genberg J Henne S and Den-tener F Better constraints on sources of carbonaceous aerosolsusing a combined 14C ndash macro tracer analysis in a Europeanrural background site Atmos Chem Phys 11 5685ndash5700httpsdoiorg105194acp-11-5685-2011 2011

Gilardoni S Massoli P Giulianelli L Rinaldi M PaglioneM Pollini F Lanconelli C Poluzzi V Carbone S HillamoR Russell L M Facchini M C and Fuzzi S Fog scav-enging of organic and inorganic aerosol in the Po Valley At-mos Chem Phys 14 6967ndash6981 httpsdoiorg105194acp-14-6967-2014 2014

Gilardoni S Massoli P Paglione M Giulianelli L Carbone CRinaldi M Decesari S Sandrini S Costabile F Gobbi G PPietrogrande M C Visentin M Scotto F Fuzzi S and Fac-chini M C Direct observation of aqueous secondary organicaerosol from biomass-burning emissions P Natl A Sci USA113 10013ndash10018 httpsdoiorg101073pnas16022121132016

Giovannini L Antonacci G Zardi D Laiti L and PanzieraL Sensitivity of Simulated Wind Speed to Spatial Reso-lution over Complex Terrain Energy Proced 59 323ndash329httpsdoiorg101016jegypro201410384 2014

Giovannini L Laiti L Serafin S and Zardi D The ther-mally driven diurnal wind system of the Adige Valley inthe Italian Alps Q J Roy Meteor Soc 143 2389ndash2402httpsdoiorg101002qj3092 2017

Gohm A Harnisch F Vergeiner J Obleitner F SchnitzhoferR Hansel A Fix A Neininger B Emeis S and SchaumlferK Air Pollution Transport in an Alpine Valley Results FromAirborne and Ground-Based Observations Bound-Lay Meteo-rol 131 441ndash463 httpsdoiorg101007s10546-009-9371-92009

Golzio A and Pelfini M High resolution WRF over mountainouscomplex terrain testing over the Ortles Cevedale area (CentralItalian Alps) in Conference First National Congress Italian As-sociation of Atmospheric Sciences and Meteorology (AISAM)AISAM 2018

Golzio A Ferrarese S Cassardo C Diolaiuti G A and PelfiniM Grid-size comparison of WRF model over complex moun-tainous terrain with topography and land-use improvementsBound-Lay Meteorol submission ID BOUND-D-19-001252019

Gong W Stroud C and Zhang L Cloud Processing of Gasesand Aerosols in Air Quality Modeling Atmosphere 2 567ndash616httpsdoiorg103390atmos2040567 2011

Green D C Fuller G W and Baker T Development and val-idation of the volatile correction model for PM10 ndash An empir-ical method for adjusting TEOM measurements for their lossof volatile particulate matter Atmos Environ 43 2132ndash2141httpsdoiorg101016jatmosenv200901024 2009

Hann J V Zur Theorie der Berg- und Talwinde Z Oumlst Meteorol14 444ndash448 1879

Harrison R M Jones A M Gietl J Yin J and Green D CEstimation of the Contributions of Brake Dust Tire Wear andResuspension to Nonexhaust Traffic Particles Derived from At-mospheric Measurements Environ Sci Technol 46 6523ndash6529 httpsdoiorg101021es300894r 2012

Hashimoto M Nakajima T Dubovik O Campanelli M CheH Khatri P Takamura T and Pandithurai G Develop-ment of a new data-processing method for SKYNET sky ra-diometer observations Atmos Meas Tech 5 2723ndash2737httpsdoiorg105194amt-5-2723-2012 2012

Hsu Y-K Holsen T M and Hopke P K Comparison ofhybrid receptor models to locate PCB sources in ChicagoAtmos Environ 37 545ndash562 httpsdoiorg101016S1352-2310(02)00886-5 2003

Kabashnikov V P Chaikovsky A P Kucsera T L and Me-telskaya N S Estimated accuracy of three common tra-jectory statistical methods Atmos Environ 45 5425ndash5430httpsdoiorg101016jatmosenv201107006 2011

Kaiser A Origin of polluted air masses in the Alps An overviewand first results for MONARPOP Environ Pollut 157 3232ndash3237 httpsdoiorg101016jenvpol200905042 2009

Kambezidis H and Kaskaoutis D Aerosol climatology over fourAERONET sites An overview Atmos Environ 42 1892ndash1906httpsdoiorg101016jatmosenv200711013 2008

Kastendeuch P P and Kaufmann P Classification ofsummer wind fields over complex terrain Int J Cli-matol 17 521ndash534 httpsdoiorg101002(SICI)1097-0088(199704)175lt521AID-JOC143gt30CO2-Q 1997

Kazadzis S Kouremeti N Dieacutemoz H Groumlbner J ForganB W Campanelli M Estelleacutes V Lantz K Michalsky JCarlund T Cuevas E Toledano C Becker R Nyeki SKosmopoulos P G Tatsiankou V Vuilleumier L Denn FM Ohkawara N Ijima O Goloub P Raptis P I Mil-ner M Behrens K Barreto A Martucci G Hall EWendell J Fabbri B E and Wehrli C Results from theFourth WMO Filter Radiometer Comparison for aerosol opti-cal depth measurements Atmos Chem Phys 18 3185ndash3201httpsdoiorg105194acp-18-3185-2018 2018

Keiser D Lade G and Rudik I Air pollution andvisitation at US national parks Sci Adv 4 7httpsdoiorg101126sciadvaat1613 2018

Khan M Masiol M Formenton G Di Gilio A de Gen-naro G Agostinelli C and Pavoni B CarbonaceousPM25 and secondary organic aerosol across the Veneto

region (NE Italy) Sci Total Environ 542 172ndash181httpsdoiorg101016jscitotenv201510103 2016

Khatri P and Takamura T An Algorithm to Screen Cloud-Affected Data for Sky Radiometer Data Analysis J Meteo-rol Soc Jpn 87 189ndash204 httpsdoiorg102151jmsj871892009

Klett J D Lidar inversion with variable backscat-terextinction ratios Appl Opt 24 1638ndash1643httpsdoiorg101364AO24001638 1985

Kukkonen J Sokhi R Luhana L Haumlrkoumlnen J Salmi T SofievM and Karppinen A Evaluation and application of a statisti-cal model for assessment of long-range transported proportion ofPM25 in the United Kingdom and in Finland Atmos Environ42 3980ndash3991 httpsdoiorg101016jatmosenv2007020362008

Landi T Curci G Carbone C Menut L Bessagnet B Giu-lianelli L Paglione M and Facchini M Simulation of size-segregated aerosol chemical composition over northern Italy inclear sky and wind calm conditions Atmos Res 125ndash126 1ndash11 httpsdoiorg101016jatmosres201301009 2013

Largeron Y and Staquet C Persistent inversiondynamics and wintertime PM10 air pollution inAlpine valleys Atmos Environ 135 92ndash108httpsdoiorg101016jatmosenv201603045 2016

Larsen B Gilardoni S Stenstroumlm K Niedzialek J JimenezJ and Belis C Sources for PM air pollution in the Po PlainItaly II Probabilistic uncertainty characterization and sensitivityanalysis of secondary and primary sources Atmos Environ 50203ndash213 httpsdoiorg101016jatmosenv201112038 2012

Loomis D Grosse Y Lauby-Secretan B El Ghissassi F Bou-vard V Benbrahim-Tallaa L Guha N Baan R MattockH and Straif K The carcinogenicity of outdoor air pollu-tion Lancet Oncol 14 1262 httpsdoiorg101016S1470-2045(13)70487-X 2013

Mann H B and Whitney D R On a Test of Whether one of TwoRandom Variables is Stochastically Larger than the Other AnnMath Stat 18 50ndash60 1947

Matta E Facchini M C Decesari S Mircea M CavalliF Fuzzi S Putaud J-P and DellrsquoAcqua A Mass clo-sure on the chemical species in size-segregated atmosphericaerosol collected in an urban area of the Po Valley Italy At-mos Chem Phys 3 623ndash637 httpsdoiorg105194acp-3-623-2003 2003

Mazzola M Lanconelli C Lupi A Busetto M Vitale V andTomasi C Columnar aerosol optical properties in the Po Val-ley Italy from MFRSR data J Geophys Res 115 D17206httpsdoiorg1010292009JD013310 2010

Meacutelin F and Zibordi G Aerosol variability in the Po Valley ana-lyzed from automated optical measurements Geophy Res Lett32 L03810 httpsdoiorg1010292004GL021787 2005

Norris G and Duvall R EPA Positive Matrix Factorization (PMF)50 ndash Fundamentals and User Guide US Environmental Protec-tion Agency available at httpswwwepagovsitesproductionfiles2015-02documentspmf_50_user_guidepdf (last access2 August 2019) 2014

Nyeki S Eleftheriadis K Baltensperger U Colbeck I FiebigM Fix A Kiemle C Lazaridis M and Petzold A AirborneLidar and in-situ Aerosol Observations of an Elevated LayerLeeward of the European Alps and Apennines Geophys Res

Lett 29 33-1ndash33-4 httpsdoiorg1010292002GL0148972002

Paatero P Least squares formulation of robust non-negativefactor analysis Chemometr Intell Lab 37 23ndash35httpsdoiorg101016S0169-7439(96)00044-5 1997

Paatero P and Tapper U Positive matrix factorization Anon-negative factor model with optimal utilization of er-ror estimates of data values Environmetrics 5 111ndash126httpsdoiorg101002env3170050203 1994

Patashnick H and Rupprecht E G Continuous PM-10Measurements Using the Tapered Element OscillatingMicrobalance J Air Waste Manage 41 1079ndash1083httpsdoiorg10108010473289199110466903 1991

Pepin N Bradley R S Diaz H F Baraer M Caceres E BForsythe N Fowler H Greenwood G Hashmi M Z LiuX D Miller J R Ning L Ohmura A Palazzi E Rang-wala I Schoumlner W Severskiy I Shahgedanova M WangM B Williamson S N and Yang D Q Elevation-dependentwarming in mountain regions of the world Nat Clim Change5 424ndash430 httpsdoiorg101038nclimate2563 2015

Perrone M Larsen B Ferrero L Sangiorgi G De GennaroG Udisti R Zangrando R Gambaro A and Bolzacchini ESources of high PM25 concentrations in Milan Northern ItalyMolecular marker data and CMB modelling Sci Total Environ414 343ndash355 httpsdoiorg101016jscitotenv2011110262012

Philipona R Greenhouse warming and solar brightening inand around the Alps Int J Climatol 33 1530ndash1537httpsdoiorg101002joc3531 2013

Pletscher K Weiss M and Moelter L Simultaneous determi-nation of PM fractions particle number and particle size distri-bution in high time resolution applying one and the same op-tical measurement technique Gefahrst Reinhalt L 76 425ndash436 available at httpwwwgefahrstoffedegestarticlephpdata[article_id]=86622 (last access 2 August 2019) 2016

Putaud J P Van Dingenen R and Raes F Submicronaerosol mass balance at urban and semirural sites in the Mi-lan area (Italy) J Geophys Res 107 LOP 11-1ndashLOP 11-10httpsdoiorg1010292000JD000111 2002

Putaud J-P Dingenen R V Alastuey A Bauer H Birmili WCyrys J Flentje H Fuzzi S Gehrig R Hansson H Harri-son R Herrmann H Hitzenberger R Huumlglin C Jones AKasper-Giebl A Kiss G Kousa A Kuhlbusch T LoumlschauG Maenhaut W Molnar A Moreno T Pekkanen J PerrinoC Pitz M Puxbaum H Querol X Rodriguez S Salma ISchwarz J Smolik J Schneider J Spindler G ten BrinkH Tursic J Viana M Wiedensohler A and Raes F AEuropean aerosol phenomenology ndash 3 Physical and chemicalcharacteristics of particulate matter from 60 rural urban andkerbside sites across Europe Atmos Environ 44 1308ndash1320httpsdoiorg101016jatmosenv200912011 2010

Putaud J P Cavalli F Martins dos Santos S and DellrsquoAcquaA Long-term trends in aerosol optical characteristics inthe Po Valley Italy Atmos Chem Phys 14 9129ndash9136httpsdoiorg105194acp-14-9129-2014 2014

Ramanathan V Crutzen P J Kiehl J T and Rosenfeld DAerosols Climate and the Hydrological Cycle Science 2942119ndash2124 httpsdoiorg101126science1064034 2001

Regione Piemonte Air quality data available at httpwwwregionepiemonteitambienteariarilevariadayariaweb-newindexphpithome last access 2 August 2019

Rizzi C Finizio A Maggi V and Villa S Spatial-temporal analysis and risk characterisation of pesticidesin Alpine glacial streams Environ Pollut 248 659ndash666httpsdoiorg101016jenvpol201902067 2019

Rosati B Gysel M Rubach F Mentel T F Goger B PoulainL Schlag P Miettinen P Pajunoja A Virtanen A KleinBaltink H Henzing J S B Groumlszlig J Gobbi G P Wieden-sohler A Kiendler-Scharr A Decesari S Facchini M CWeingartner E and Baltensperger U Vertical profiling ofaerosol hygroscopic properties in the planetary boundary layerduring the PEGASOS campaigns Atmos Chem Phys 167295ndash7315 httpsdoiorg105194acp-16-7295-2016 2016

Saarikoski S Carbone S Decesari S Giulianelli L AngeliniF Canagaratna M Ng N L Trimborn A Facchini M CFuzzi S Hillamo R and Worsnop D Chemical characteriza-tion of springtime submicrometer aerosol in Po Valley Italy At-mos Chem Phys 12 8401ndash8421 httpsdoiorg105194acp-12-8401-2012 2012

Sabatier T Paci A Canut G Largeron Y Dabas A DonierJ-M and Douffet T Wintertime Local Wind Dynamics fromScanning Doppler Lidar and Air Quality in the Arve River Val-ley Atmosphere 9 118 httpsdoiorg103390atmos90401182018

Samset B H How cleaner air changes the climate Science 360148ndash150 httpsdoiorg101126scienceaat1723 2018

Sandrini S Fuzzi S Piazzalunga A Prati P Bonasoni P Cav-alli F Bove M C Calvello M Cappelletti D Colombi CContini D de Gennaro G Di Gilio A Fermo P Ferrero LGianelle V Giugliano M Ielpo P Lonati G Marinoni AMassabograve D Molteni U Moroni B Pavese G Perrino CPerrone M G Perrone M R Putaud J-P Sargolini T Vec-chi R and Gilardoni S Spatial and seasonal variability of car-bonaceous aerosol across Italy Atmos Environ 99 587ndash598httpsdoiorg101016jatmosenv201410032 2014

Schaap M van Loon M ten Brink H M Dentener F J andBuiltjes P J H Secondary inorganic aerosol simulations forEurope with special attention to nitrate Atmos Chem Phys 4857ndash874 httpsdoiorg105194acp-4-857-2004 2004

Schmidli J Daytime Heat Transfer Processes overMountainous Terrain J Atmos Sci 70 4041ndash4066httpsdoiorg101175JAS-D-13-0831 2013

Schmidli J Boumling S and Fuhrer O Accuracy of Simulated Diur-nal Valley Winds in the Swiss Alps Influence of Grid ResolutionTopography Filtering and Land Surface Datasets Atmosphere9 196 httpsdoiorg103390atmos9050196 2018

Seibert P The riddles of foehn ndash introduction to the his-toric articles by Hann and Ficker Meteorol Z 21 607ndash614httpsdoiorg1011270941-294820120398 2012

Seibert P Kromp-Kolb H Baltensperger U Jost D Tand Schwikowski M Trajectory Analysis of High-AlpineAir Pollution Data Springer US Boston MA 595ndash596httpsdoiorg101007978-1-4615-1817-4_65 1994

Seibert P Kromp-kolb H Kasper A Kalina M PuxbaumH Jost D T Schwikowski M and Baltensperger UTransport of polluted boundary layer air from the Po Val-

ley to high-alpine sites Atmos Environ 32 3953ndash3965httpsdoiorg101016S1352-2310(97)00174-X 1998

Seinfeld J H and Pandis S N Atmospheric Chemistry andPhysics From Air Pollution to Climate Change ndash 2nd ed JohnWiley amp Sons 2006

Serafin S and Zardi D Daytime Heat Transfer ProcessesRelated to Slope Flows and Turbulent Convection in anIdealized Mountain Valley J Atmos Sci 67 3739ndash3756httpsdoiorg1011752010JAS34281 2010

Serafin S Adler B Cuxart J De Wekker S F J GohmA Grisogono B Kalthoff N Kirshbaum D J Ro-tach M W Schmidli J Stiperski I Vecenaj Ž andZardi D Exchange Processes in the Atmospheric Bound-ary Layer Over Mountainous Terrain Atmosphere 9 102httpsdoiorg103390atmos9030102 2018

Silibello C Calori G Brusasca G Giudici A AngelinoE Fossati G Peroni E and Buganza E Modellingof PM10 concentrations over Milano urban area using twoaerosol modules Environ Modell Softw 23 333ndash343httpsdoiorg101016jenvsoft200704002 2008

Skamarock W C Evaluating Mesoscale NWP Models UsingKinetic Energy Spectra Mon Weather Rev 132 3019ndash3032httpsdoiorg101175MWR28301 2004

Sprenger M and Wernli H The LAGRANTO Lagrangian anal-ysis tool ndash version 20 Geosci Model Dev 8 2569ndash2586httpsdoiorg105194gmd-8-2569-2015 2015

Squizzato S and Masiol M Application of meteorology-based methods to determine local and external con-tributions to particulate matter pollution A casestudy in Venice (Italy) Atmos Environ 119 69ndash81httpsdoiorg101016jatmosenv201508026 2015

Stohl A Trajectory statistics-A new method to establish source-receptor relationships of air pollutants and its application to thetransport of particulate sulfate in Europe Atmos Environ 30579ndash587 httpsdoiorg1010161352-2310(95)00314-2 1996

Tampieri F Trombetti F and Scarani C Summer daily circula-tion in the Po Valley Italy Geophys Astro Fluid 17 97ndash112httpsdoiorg10108003091928108243675 1981

Tang L Haeger-Eugensson M Sjoumlberg K Wichmann JMolnaacuter P and Sallsten G Estimation of the long-rangetransport contribution from secondary inorganic componentsto urban background PM10 concentrations in south-westernSweden during 1986ndash2010 Atmos Environ 89 93ndash101httpsdoiorg101016jatmosenv201402018 2014

Thunis P Pederzoli A and Pernigotti D Performance criteria toevaluate air quality modeling applications Atmos Environ 59476ndash482 httpsdoiorg101016jatmosenv201205043 2012

Thyer N H A theoretical explanation of mountain and valleywinds by a numerical method Arch Meteor Geophy A 15318ndash348 httpsdoiorg101007BF02247220 1966

Tudoroiu M Eccel E Gioli B Gianelle D Schume H Gen-esio L and Miglietta F Negative elevation-dependent warm-ing trend in the Eastern Alps Environ Res Lett 11 044021httpsdoiorg1010881748-9326114044021 2016

Van Donkelaar A Martin R V Brauer M Kahn RLevy R Verduzco C and Villeneuve P J Global es-timates of ambient fine particulate matter concentrationsfrom satellite-based aerosol optical depth development

and application Environ Health Persp 118 847ndash855httpsdoiorg101289ehp0901623 2010

Wagner J S Gohm A and Rotach M W The impact of val-ley geometry on daytime thermally driven flows and verticaltransport processes Q J Roy Meteor Soc 141 1780ndash1794httpsdoiorg101002qj2481 2014a

Wagner J S Gohm A and Rotach M W The Impact of Hori-zontal Model Grid Resolution on the Boundary Layer Structureover an Idealized Valley Mon Weather Rev 142 3446ndash3465httpsdoiorg101175MWR-D-14-000021 2014b

Waked A Favez O Alleman L Y Piot C Petit J-E Delau-nay T Verlinden E Golly B Besombes J-L Jaffrezo J-L and Leoz-Garziandia E Source apportionment of PM10 ina north-western Europe regional urban background site (LensFrance) using positive matrix factorization and including pri-mary biogenic emissions Atmos Chem Phys 14 3325ndash3346httpsdoiorg105194acp-14-3325-2014 2014

Wang X Wang W Yang L Gao X Nie W Yu Y XuP Zhou Y and Wang Z The secondary formation of inor-ganic aerosols in the droplet mode through heterogeneous aque-ous reactions under haze conditions Atmos Environ 63 68ndash76httpsdoiorg101016jatmosenv201209029 2012

Weissmann M Braun F J Gantner L Mayr G J RahmS and Reitebuch O The Alpine MountainndashPlain Cir-culation Airborne Doppler Lidar Measurements and Nu-merical Simulations Mon Weather Rev 133 3095ndash3109httpsdoiorg101175MWR30121 2005

WHO Ambient air pollution a global assessment of exposure andburden of disease Tech rep 2016

Wiegner M and Geiszlig A Aerosol profiling with the Jenop-tik ceilometer CHM15kx Atmos Meas Tech 5 1953ndash1964httpsdoiorg105194amt-5-1953-2012 2012

WMO WMOIGAC Impacts of megacities on air pollution andclimate Tech rep World Meteorological Organization avail-abel at httpslibrarywmointpmb_gedgaw_205pdf (last ac-cess 2 August 2019) 2012

WMO WMO Global Atmosphere Watch (GAW) Implementa-tion Plan 2016-2023 Tech rep World Meteorological Or-ganization available at httpslibrarywmointdoc_numphpexplnum_id=3395 (last access 2 August 2019) 2017

Wotawa G Kroumlger H and Stohl A Transport of ozone to-wards the Alps ndash results from trajectory analyses and pho-tochemical model studies Atmos Environ 34 1367ndash1377httpsdoiorg101016S1352-2310(99)00363-5 2000

Ying C Chunsheng Z Qiang Z Zhaoze D Mengyu Hand Xincheng M Aircraft study of Mountain ChimneyEffect of Beijing China J Geophys Res 114 D08306httpsdoiorg1010292008JD010610 2009

Yuan Y Ries L Petermeier H Steinbacher M Goacutemez-PelaacuteezA J Leuenberger M C Schumacher M Trickl T CouretC Meinhardt F and Menzel A Adaptive selection of diur-nal minimum variation a statistical strategy to obtain represen-tative atmospheric CO2 data and its application to European el-evated mountain stations Atmos Meas Tech 11 1501ndash1514httpsdoiorg105194amt-11-1501-2018 2018

Zardi D and Whiteman C D Diurnal Mountain WindSystems Springer Netherlands Dordrecht 35ndash119httpsdoiorg101007978-94-007-4098-3_2 2013

Zeng Y and Hopke P A study of the sources of acid precip-itation in Ontario Canada Atmos Environ 23 1499ndash1509httpsdoiorg1010160004-6981(89)90409-5 1989

Zeng Z Chen A Ciais P Li Y Li L Z X Vautard RZhou L Yang H Huang M and Piao S Regional airpollution brightening reverses the greenhouse gases inducedwarming-elevation relationship Geophys Res Lett 42 4563ndash4572 httpsdoiorg1010022015GL064410 2015

Zong Z Wang X Tian C Chen Y Fu S Qu L Ji L Li Jand Zhang G PMF and PSCF based source apportionment ofPM25 at a regional background site in North China Atmos Res203 207ndash215 httpsdoiorg101016jatmosres2017120132018

Zuev V V Burlakov V D Nevzorov A V Pravdin V LSavelieva E S and Gerasimov V V 30-year lidar obser-vations of the stratospheric aerosol layer state over Tomsk(Western Siberia Russia) Atmos Chem Phys 17 3067ndash3081httpsdoiorg105194acp-17-3067-2017 2017

Abstract

Introduction

Study region and experimental setup

Modelling tools

Numerical atmospheric models

Trajectory statistical models

Positive matrix factorisation

Results

Daily classification schemes based on ALC measurements or meteorology

Vertical and temporal characteristics of the advected aerosol layer

Impact of Po Valley advections on northwestern Alps surface-level PM loads and physico-chemical characteristics

Surface PM variability in relation to the ALC profiles classification scheme

Chemical species within the advected aerosol layers

PMF of aerosol size distributions and links to chemical properties

Impact of Po Valley advections on northwestern Alps columnar aerosol properties

Comparison between observations and CTM simulations

Conclusions and perspectives

Data availability

Supplement

Author contributions

Competing interests

Special issue statement

Acknowledgements

Review statement

References