Application of the Lidar/photometer algorithm: dust model ...Manuscript prepared for Atmos. Meas. Tech. with version 2014/07/29 7.12 Copernicus papers of the LATEX class coperni- cus.cls

Manuscript prepared for Atmos. Meas. Tech.with version 2014/07/29 7.12 Copernicus papers of the LATEX class coperni-cus.cls.Date: 4 December 2014

Application of the Lidar/photometer algorithm: dust modelevaluationI. Binietoglou1, S. Basart2, V. Amiridis3, L. A. Arboledas4, A. Argyrouli5, J. M. Baldasano2, D. Balis6, L. Belegante1, J.A. Bravo-Aranda4, P. Burlizzi7, V. Carrasco8, A. Chaikovsky9, A. Comerón10, G. D’Amico11, M. Filioglou6, M. J.Granados-Muños4, J. L. Guerrero-Rascado4, P. Kokkalis5, A. Maurizi12, F. Monti12, C. Muñoz10, D. Nicolae1, S.Nikcovic13, A. Papayannis5, G. Pappalardo11, S. N. Pereira8, M. R. Perrone7, A. Pietruczuk14, M. Posyniak14, A.Rodriguez10, M. Sicard10, 14, N. Siomos6, A. Szkop14, E. Terradellas16, A. Tsekeri3, U. Wandinger17, and J. Wagner17

1National Institute of R&D for Optoelectronics, 409 Atomistilor Str., 77125, Magurele, Ilfov, Romania2Earth Sciences Department, Barcelona Supercomputing Center-Centro Nacional de Supercomputación, BSC-CNS,Barcelona, Spain3National Observatory of Athens, Institute for Astronomy, Astrophysics, Space Applications and Remote Sensing(NOA-IAASARS), Athens, Greece4Departamento de Física Aplicada, Universidad de Granada, Granada, Spain5National Technical University of Athens, Physics Department, Laser Remote Sensing Laboratory, Zografou, Greece6Aristotle University of Thessaloniki, Faculty of sciences, School of physics, 54124, Thessaloniki, Greece7Dipartemento di Fisica, Universitá di Lecce, Lecce, Italy8Èvora Geophysics Centre, Èvora, Portugal9Institute of Physics, National Academy of Sciences of Belarus, Minsk, Belarus10Department of Signal Theory and Communications, Remote Sensing Laboratory, Universitat Politècnica de Catalunya,Barcelona, Spain11Consiglio Nazionale delle Ricerche, Istituto di Metodologie per l’Analisi Ambientale (CNR-IMAA), Tito Scalo, Potenza,Italy12Consiglio Nazionale delle Ricerche, Istituto di Scienze dell’Atmosfera e del Clima (CNR-ISAC), Bologna, Italy13Institute of Physics, Belgrade, Serbia14Atmospheric Physics Division, Institute of Geophysics University of Warsaw, Warszawa, Poland15IEEC-CRAE, Universitat Politècnica de Catalunya, Barcelona, Spain16AEMET, Barcelona, Spain17Leibniz Institute for Tropospheric Research Leipzig, Germany18Faculty of Agriculture, University of Belgrade, Serbia

Correspondence to: I.Binietoglou([email protected])

Abstract.Systematic measurement of dust concentration profiles at

continental scale was recently made possible by the develop-ment of sophisticated retrieval algorithms based on the syn-ergistic use of combined lidar and sun photometer data and5

the establishment of robust remote sensing networks in theframework of ACTRIS/EARLINET. In this paper, we presenta methodology for using these capabilities as a tool for theevaluation of dust transport models. The methodology in-cludes considerations for the selection of a suitable dataset10

and appropriate metrics for the exploration of the results. Theevaluation approach is demonstrated for four regional dusttransport models (BSC/DREAM8bV2, BSC/NMMB-DUST,DREAMABOL, DREAM8-NMME-MACC) using dust ob-servations performed at 10 ACTRIS/EARLINET stations.15

The observations include coincident multi-wavelength lidarand sun photometer measurements, while the concentrationretrievals were done with the utilization of the LIRIC al-gorithms. The dust model evaluation methodology proposedhere shows advantages when compared to traditional valida-20

2 I.Binietoglou et al.: Dust transport model evaluation

tion techniques that utilize separately the available measure-ments; these include the separation of the fine and coarse par-ticle mode on the lidar profiles and the avoidance of modelassumptions related to the conversion of concentration fieldsto aerosol extinction values. When compared to LIRIC re-25

trievals, the simulated dust vertical structures of dust werefound to be in good agreement for all models, while the ab-solute dust concentration was typically underestimated. Thereported differences among the models found in this compar-ison indicate the benefit of the systematic use of the proposed30

approach in future dust model evaluation studies.

1 Introduction

Desert dust is produced in arid regions around the world,and in many cases they are the dominant aerosol type. Dustaerosols affect the radiation balance and temperature struc-35

ture of the atmosphere by interacting both with short- andlong-wave radiation (Sokolik and Toon, 1996); they also af-fect cloud micro-physical properties and precipitation pat-terns by acting as could and ice condensation nuclei and, dueto their large spatial and temporal extent, have important ef-40

fect on climate (Rosenfeld et al., 2001). The main source re-gions of dust are located in Northern Africa and central Asiabut due to the prevalent wind patterns they have significantimpact on the air quality of Europe, North America and EastAsia, far away from their source. Additionally, mineral dust45

aerosols are suspected to be a important source of solubleiron in the marine ecosystems and, thus, an important factorof marine bioproduction (Nickovic et al., 2013).

Given this complexity, dust transport models are an im-portant tool for studying the complete dust cycle in the at-50

mosphere. Such models simulate dust’s life-cycle includingproduction in arid regions, transport in the atmosphere, andwet and dry deposition. These models can produce complete3D fields of dust distribution, can be used to provide dustconcentration forecasts, and, in total, give a birds-eye view55

of the atmosphere. To perform these simulation, models relyon physical description of atmospheric processes, choice ofparameterization, and the tuning of individual component ofthe model; consequently, modeling output need to be mon-itored against in situ and remote sensing measurements to60

try to evaluate their performance. When used as a forecasttool, remote sensing measurements can be assimilated intothe model to improve the forecasting skill of the models(Benedetti et al., 2009; Sekiyama et al., 2010).

Most dust model evaluation methods are based on colum-65

nar measurements of total-column aerosol properties. Typ-ical quantities used, for example, are aerosol optical depth(AOD) measured by the AERONET photometer network orsatellite platforms, such as MODIS (Pérez et al., 2011; Basartet al., 2012). In these comparisons, the dust volume concen-70

tration is converted to optical depth using the spherical par-

ticle approximation and a modeled size distribution. Theseevaluation attempts are limited by the contribution of non-dust aerosols and so are restricted in cases or regions thatdust is the dominant aerosol type.75

While the columnar properties of dust are systematicallystudied, the vertical distribution of dust is a parameter thathas not been explored in the same extent, even though thiscould have a significant effect in the total model’s perfor-mance. A better vertical distribution, for example, could have80

significant effect in the transport and removal component ofthe dust model and would have significant impact on thequality of the air-quality forecasts and the study of dust-radiation and dust-cloud interactions.

In this paper, we examine a strategy for evaluating the85

dust’s vertical distribution at a continental scale. Specifically,we demonstrate how advanced mass concentration retrievalalgorithm and the establishment of robust, quality assured,remote sensing networks can be used for the systematic eval-uation of regional dust models. The use of the advanced re-90

trieval algorithms permit the direct evaluation of the simu-lated dust concentration without relying on the, highly un-certain, model’s dust volume distribution and Mie code cal-culations. The use of continental-scale networks permit theevaluation of the vertical distribution of dust in different me-95

teorological conditions and measurement locations, provid-ing a more complete view of model’s performance.

The vertical distribution of dust has been previously stud-ied using lidar optical property profiles, but these stud-ies have focused at single locations or in few case stud-100

ies. For example, Mona et al. (2014) have presented a sys-tematic evaluation of BSC/DREAM8b model vertical dis-tribution over Potenza, Italy, using lidar-derived backscatterand extinction profiles. Similarly, Gobbi et al. (2013) com-pare the lidar dust extinction profiles with those modeled105

by BSC/DREAM8b over Rome, Italy during the 2001-2004period. Their results indicate that the model represent ade-quately the vertical distribution of dust but find the total ex-tinction profiles underestimated.

The development of advanced algorithms allows the re-110

trieval of dust concentration profiles using synergy of lidarand sun/sky photometer data (e.g. Chaikovsky (2015); Ans-mann et al. (2012); Lopatin et al. (2013)) and the quanti-ties retrieved by the remote sensing measurements can bedirectly compared to the dust model’s output. Using such al-115

gorithms, the contribution of dust is separated from that ofother aerosol types and the comparison can be extended alsoin mixed aerosol cases, even at those with low dust concen-tration. Additionally, the retrieved products include informa-tion about the actual aerosol size distribution — instead of120

relying on the model simulated size distribution — furtherimproving the results. Up to now, the comparison of thesealgorithms with model’s has been restricted in single cases;for example, Tsekeri et al. (2013) present a case study wherethe output of DREAM8b model is compared with dust con-125

I.Binietoglou et al.: Dust transport model evaluation 3

centration retrieved using the LIRIC algorithm, over Athens,Greece, finding satisfactory agreement.

Recently, LIRIC has been implemented in many advancedEuropean remote sensing stations (Chaikovsky et al., 2012)allowing the systematic comparison of model’s performance130

in a wider geographical region. In this paper we present ageneral methodology for the measurement-model compari-son of the vertical dust distribution including the strategiesthat could be used, the caveats that should be taken care of,and suggest the appropriate metrics that could help explore135

the dataset. Next, we apply this methodology to compare dustconcentration profiles retrieved at 10 European remote sens-ing sites to 4 regional dust transport models that are part ofWMO’s SDS-WAS.

The rest of the paper is structured as follows: In sec-140

tion 2 we present remote sensing networks EARLINET andAERONET, we provide an overview of the new retrieval al-gorithms together with cases that their use can be advanta-geous to dust model evaluation, and present the four dustmodels used in this study. In section 3 we present the method-145

ology of the comparison, and examine the appropriate met-rics that can be used for future evaluation of dust models.In section 4 we present the results obtained by applying thismethodology to real measurement. In section 5 we give con-clusions and indicate direction for future work.150

2 Algorithms & Models

2.1 Measurement networks

The systematic observation of the vertical distribution of duston continental scale is possible due to the development ofregional lidar remote sensing networks in main dust out-155

flow regions like the European Aerosol Research Lidar Net-work (EARLINET) (Pappalardo et al., 2014), the AD-Net inEast Asia (Sugimoto et al., 2005), and the ALINE networkis Latin America (Barbosa et al., 2014). This study focuseson EARLINET, a lidar network that has been established160

in 2000 with the aim to provide comprehensive informationfor the aerosol distribution over Europe (Bösenberg et al.,2001). Currently, 27 stations participate activly in the net-work with regular contribution of data. There are currently18 stations with multiwavelength Raman systems, while 17165

stations currently perform depolarization measurements, thatgive important information on the shape of the measuredparticles. All stations in the network perform climatologicalmeasurements — three times a week according to a prede-fined measurement schedule — together with extra measure-170

ments in special events, like dust measurements based on analerting system, and intense measurement campaigns (Pap-palardo et al., 2014). Considerable attention has been givenwithin EARLINET to improve and homogenize the perfor-mance of the systems, including hardware test, algorithm175

test on synthetic data and system intercomparison campaigns

(Matthias et al., 2004; Böckmann et al., 2004; Pappalardoet al., 2004). The optical products calculated from all thesystems are stored in a standardized data format in a cen-tral database and are available for external users. The first180

volumes of the EARLINET database have been published inbiannual volumes at the World Data Center for Climate. (TheEARLINET publishing group 2000-2010 et al., 2014)

Similarly, regional-to-global sun/sky photometer networkslike Aerosol Robotic Network (AERONET) and Skyrad Net-185

work (SKYNET) have also been developed (Holben et al.,1998; Takamura and Nakajima, 2004; Kim et al., 2008);many of these instruments are collocated with lidar system ofthe corresponding lidar networks allowing the developmentof synergistic algorithms. In this study we use the Aerosol190

Robotic Network (AERONET), a global network of auto-matic sun/sky-scanning photometers that was created in themid 90’s in order to provide global aerosol data not pro-vided at the time by satellites and to act as a validation plat-form for future satellite missions. Its current aim is to pro-195

vide long-term, continuous measurements of aerosol opti-cal and microphysical properties. The network consists ofstandardized photometers produced by Cimel Electroniqueand all participating instruments undergo regular calibrationand intercomparison with reference instruments. The pho-200

tometers in the AERONET network perform both direct-sunand sky-scanning almucantar measurements. The main out-put of the former is the aerosol optical depth in several wave-lengths, while the later are used to retrieve aerosol micro-physical properties (Dubovik and King, 2000; Dubovik et al.,205

2006). The processing is centrally performed and the resultsare made publicly available in near real time.

2.2 Volume retrieval algorithms

Different levels of lidar-based remote sensing products canbe used for the evaluation of dust models, ranging from un-210

calibrated range-corrected signals (RCS) to dust mass con-centration profiles retrieved through synergistic algorithms;an overview of the available lidar products is given in Table1. The first level of products that can be used for model evalu-ation are range-corrected lidar signals that give qualitatively215

information about the aerosol structure in the atmosphere;they can give an overview of dust transport processes andhelp check the geometrical properties of the simulated dustlayers. Being almost raw lidar products, RCS can be providedeven by simple lidar systems and some ceilometers (Wiegner220

et al., 2014). On a second level, retrieved optical properties,i.e. profiles of aerosol backscatter and extinction coefficients,give quantitative information about the total aerosol content;these products can be used to study both geometrical and in-tensive properties of the dust layers and are especially use-225

ful in strong lofted dust layers. They can not be used reli-ably, however, in cases of mixed aerosol, as they don’t sepa-rate the contribution of dust from other atmospheric aerosol,like smoke and pollution. If depolarization measurements are


available, this problem is solved by a third level of analysis,230

that retrieve the dust backscatter coefficient profiles, basedon known depolarization ratios of dust and other types ofaerosols particles assumed in the atmosphere; this allows thedirect comparison of modeled dust backscatter profiles withthe measured ones, without the biases introduced by other235

aerosol mixtures.A fourth level of products has been developed in recent

years, motivated by an increased interest in extracting aerosolconcentration profiles from remote sensing measurements.Several algorithms have been developed, combining the ver-240

tically resolved lidar measurements with photometer data orassumed aerosol extensive properties; the output of these al-gorithms is the vertical concentration of a number of sep-arate aerosol types. The use of these algorithms addressesa core issue of model evaluation from remote sensing mea-245

surements: dust transport models simulate mass concentra-tion while the main measured quantities of remote sensinginstruments are optical aerosols properties; a conversion isalways necessary to make the two quantities comparable. Thecomparison is typically done by converting the model’s mass250

concentration to extinction profiles using an assumed or sim-ulated volume-to-extinction ratio. If the dust transport modeltreats the dust size distribution in a realistic way, e.g. sepa-rating the dust concentration in many different size bins, abetter conversion can be achieved using forward scattering255

calculations (typically Mie approximation) on the simulatedsize distribution. In contrast, when using the remote sensingalgorithms presented before, the retrieved quantities can bedirectly compared to the model’s output. The main benefitof this comparison is that dust microphysical properties are260

neither assumed a priori nor are based on model outputs, butare estimated using actual photometer measurements or mea-surements of pure dust types.

An additional benefit of using concentration retrieval algo-rithms, is that the contribution of dust can be separated from265

the one of other aerosol types present in the atmosphere, e.g.from urban or smoke aerosols. In contrast, when using di-rect measured lidar products, like backscatter and extinctionprofiles, other aerosol types can have important contributionto the measured profiles, and this restricts the comparison to270

assumed ”pure” dust cases. The comparison should be alsorestricted in these cases above the Planetary Bounday Layer(PBL), as the load of fine anthropogenic aerosol is the PBL isalways expected, especially in most European measurementsites.275

A point that should be taken into account when choosingthe appropriate remote sensing product for dust model vali-dation, is the required vertical resolution. Specifically, somedust related products could have a lower effective verticalresolution compared to the backscatter and extinction pro-280

files of lidar. For most application this vertical resolution isstill higher than the one from the dust models, so it shouldnot affect the results of the comparison. A case that the opti-cal property profiles could be preferable is when determining

the geometrical characteristics, i.e. top and bottom altitude,285

of the dust layer. In these cases, the backscatter lidar productsare more appropriate.

The existing mass retrieval algorithms can be separated intwo broad categories. The first category uses lidar measure-ments and intensive optical properties of some aerosol types290

to retrieve the concentration of these types in the atmosphere.The used aerosol intensive properties can be derived frompast observations, laboratory measurements, model data or acombination of the above. When the range of a needed valueis too wide for reliable retrieval, photometer measurement295

are sometime used as a proxy for the missing parameter. Forexample, the POLIPHON algorithm (Ansmann et al., 2011,2012) is based on dust depolarization and lidar ratio prop-erties derived during the SAMUM campaign and long-termEARLINET measurements of dust transport events over Eu-300

rope. In addition, POLIPHON uses the volume-to-AOD ra-tio derived from the photometer to approximate the variablevolume-to-extinction ratio for dust and smoke aerosol. Ex-tending this approach, Mamouri and Ansmann (2014) uselaboratory measurements of fine and coarse dust depolariza-305

tion ratio to further separate these two sub-classes of dust. Incontrast, Nemuc et al. (2013) use properties from the OPACoptical property database (Hess et al., 1998) as a basis fortheir separation of dust and non-dust properties. Other ap-proaches combining lidar measurements with airborne mea-310

surements or complex AERONET processing have also beendeveloped (Cuesta et al., 2008; Lewandowski et al., 2010).

The second category of algorithms pursues a more tightintegration of lidar and photometer data. Specifically, the re-trieved volume concentration profiles are calculated by opti-315

mally fitting the lidar and sky-scanning photometer measure-ments (Dubovik, 2005). In the case of GARRLIC (Lopatinet al., 2013), the optimal fit of the lidar and photometer mea-surements is found using a multi-term least square approach.Similarly, LIRIC (Chaikovsky, 2015) uses the AERONET in-320

version products to derive the intensive properties of fine andcoarse aerosols; consequently the algorithm finds the optimalprofile of these types based on lidar and total-column volumeconcentration profiles provided by AERONET.

In this paper, we use results from the LIRIC algorithms325

to show the benefit of using such algorithms for dust modelevaluation. LIRIC was chosen as it takes full advantage ofremote sensing networks EARLINET and AERONET, andis used by a big number of aerosol remote sensing station inEurope (Chaikovsky et al., 2012); the results we present are,330

nevertheless, applicable to similar datasets retrieved by otheralgorithms. The details of LIRIC can be found in Chaikovsky(2015) so only a brief overview is given here.

LIRIC assumes that the aerosols in the atmosphere, atany given moment, can be described as a mixture of few335

aerosol component; these components are expected to havestable microphysical properties throughout the atmosphere,but vary their concentration with altitude. LIRIC retrievesthese concentration profiles based on AERONET’s micro-


physical inversion and multiwavelenght lidar measurements.340

The fixed intensive properties of the components are calcu-lated based on AERONET’s microphysical inversion. Specif-ically, the retrieved aerosol size distribution is split intofine and coarse modes by finding its minimum value in the0.194 - 0.576 µm range. The intensive properties of the345

two modes are calculated assuming a mixture of spheresand randomly oriented spheroids. When lidar depolarizationmeasurements are available, the coarse mode can be fur-ther separated into spherical and spheroidal part, based onAERONET’s sphericity parameter. Finally, LIRIC retrieves350

the component’s volume concentration profiles that simul-taneously optimize the fit to lidar backscatter signals andto columnar volume concentration values retrieved by thephotometer. In this way the volume concentration of fine,coarse, and, possibly, coarse/non-spherical aerosol modes are355

retrieved.LIRIC has proven to be a robust algorithm for aerosol vol-

ume concentration retrievals. The output of LIRIC has beenvalidated against similar retrievals that don’t rely on a spe-cific aerosol model (Wagner et al., 2013); the comparison in-360

dicates that the spheroid model that represent non sphericalparticles does not induce significant errors in the retrieval. Afurther source of uncertainties is the choice of user-definedparameters for each retrieval; such parameters include, forexample, minimum and maximum altitude, the altitude of365

an aerosol-free region, and regularization parameters used inthe inversion. Granados-Muñoz et al. (2014) show that theretrieval is stable to the choice of these parameters; in the ex-amples shown in that paper, the measurement error remainsbelow 20%.370

2.3 The WMO SDS-WAS program and the dust forecastmodels

Dust transport modeling was the point of intense researchduring the last decades and several global and regional mod-els have been developed (Tegen and Fung, 1994; Nickovic375

and Dobricic, 1996; Benedetti et al., 2014). These modelsinclude soil type data to describe dust sources and are drivenby meteorological field of global atmospheric models. Therepresentation of sources, the meteorological driver, the de-position scheme, the handling of dust of different size ranges,380

and other choice of parameters create very different behaviorof each dust model.

In this study we focus on regional transport models setupover the domain of North Africa and Europe; these modelsare frequently used to predict dust transport over Europe and385

to explore the effects of dust in the European atmosphere.Such regional dust models have been used, for example, toquantify the effect of dust on air quality of Mediterraneancities (Jiménez-Guerrero et al., 2008), to study the effects ofdust on weather forecasts (Pérez et al., 2006b), and to quan-390

tify the impact of lofted dust particles on cloud formation

(Solomos et al., 2011); correctly representing the vertical ex-tent of dust is a crucial aspect of all these studies.

The four models that participate in this inter-comparisonare BSC-DREAM8bV2, NMMB/BSC-Dust, DREAM-395

ABOL, and DREAM8-NMME-MACC; an overview oftheir setup is given in Table 2. The four models con-tribute to the Sand and Dust Storm Warning Advisoryand Assessment System (SDS-WAS) Program that wasestablished by the World Meteorological Organization400

(http://www.wmo.int/sdswas). The SDS-WAS Program aimsto improve the present capabilities for reliable sand anddust storm forecasts; to do this it supports the developmentof comprehensive, coordinated and sustained observationsand modeling capabilities of these events. The SDS-WAS405

Program consists of two Regional Nodes, one for NorthernAfrica, Middle East and Europe (NA-ME-E) – hosted inSpain – and one in Asia – hosted in China; each of thesenodes deals with both operational and scientific aspectsrelated to atmospheric dust monitoring and forecasting.410

All models examined in this study undergo near real timeevaluation against satellite- and ground-based columnarobservations.

The basis of three out of four examined models is the DustRegional Atmospheric Model (DREAM) (Nickovic et al.,415

2001). Originally, DREAM was based on the Euler-type par-tial differential nonlinear equation for dust mass continuityand is driven by NCEP/Eta model (Janjic, 1994). The de-velopment of the model started at the Euro-MediterraneanCentre of Insular Coastal Dynamics (ICoD) and later moved420

to the Environmental Modelling Laboratory of the TechnicalUniversity of Catalonia (UPC) and the Barcelona Supercom-puting Center (BSC-CNS). The version of the model usedin this study (BSC/BSC-DREAM8b v2) was subject to sev-eral updates including eight particle-size bins representation425

of the dust size distribution, improved source representation,and updated wet and dry deposition schemes (Pérez et al.,2006b, a; Basart et al., 2012).

The NMMB/BSC-Dust model is a regional to global dustforecast system designed and developed at BSC in collabora-430

tion with NOAA NCEP, NASA Goddard Institute for SpaceStudies and the International Research Institute for Climateand Society (IRI) (Pérez et al., 2011). It is an online multi-scale atmospheric dust model fully embedded into the Non-hydrostatic Multiscale Model (NMMB) (Janjic et al., 2011).435

As with DREAM, this model also utilizes an 8-bin size dis-tribution while it includes a physically based dust emissionscheme, which explicitly takes into account saltation andsandblasting processes. The NMMB/BSC-Dust model hasbeen evaluated at regional and global scales (Pérez et al.,440

2011; Haustein et al., 2012).The DREAMABOL model is a online integrated regional

mineral dust model developed at the Institute of AtmosphericSciences and Climate, Bologna, Italy, as part of the atmo-spheric composition and meteorology model BOLCHEM445

(Mircea et al., 2008; Maurizi et al., 2011). The meteorologi-


cal component is the BOLAM primitive equation hydrostaticmodel (Buzzi et al., 2003). Dust model part is inspired byDREAM (Nickovic et al., 2001) but is completely rewrittenand includes different assumptions on the model source and450

on the wet removal (Maurizi and Monti). DREAMABOL en-tered the SDS-WAS community on June 2014 and partici-pates since then in the near real time evaluation.

The DREAM8-NMME-MACC is developed and operatedat the South East European Virtual Climate Change Center455

(SEEVCCC), Serbia. The model is embedded in the NCEPNonhydrostatic Mesoscale Model on E-grid (Janjic et al.,2001) while initial and boundary conditions are taken fromECMWF global forecast. This version of DREAM8 is as-similating ECMWF dust analysis in dust initial field, with460

dust sources defined from Ginoux et al. (2001). DREAM8-NMME-MACC provides daily dust forecasts available atthe South East European Virtual Climate Change Center(SEEVCCC).

3 Methodology465

In this section we present the considerations for constructingthe remote sensing dataset and the metrics than were usedfor model evaluation purposes. Special attention is given inselecting a representative dataset, avoiding possible biasesdue to the geographical restrictions of the measurement lo-470

cation, the selection of vertical resolution, and the effect oflocal dust sources in the study of the PBL. The considerationsthat guided our choices are given below.

Remote sensing profiling measurements can be used toimprove dust modeling efforts at three different levels: di-475

agnostic evaluation, Near Real-Time (NRT) evaluation, andassimilation (Seigneur et al., 2000). In the first case, remotesensing measurements are used to study the model’s perfor-mance during a past study period. The aim of such a studyis to evaluate the model’s performance, understand its be-480

havior and limitations, and suggest improvement either bytuning applied parameterizations or changing representationof processes in the models. The evaluation can focus on in-dividual cases of dust transport events or follow a statisticalapproach, covering a larger time period. In the case of NRT485

evaluation, the measured profiles are used to provide insighton the performance of an operational forecast. The aim ofsuch evaluation is to provide quick overview of the model’sperformance to the users while it can also help modelers de-tect possible problems in time. The time requirements for the490

processing of such observations are moderately strict, as datacould be useful even one day after the measurement time. Inthe case of assimilation, the remote sensing measurementsare used by the model to improve the forecasted model con-centrations. The time requirements for assimilation depend495

on the assimilation system used but are typically few hours.In this study we focus on the fist case, the diagnostic eval-uation of model’s performance. We choose to study an ex-

tend time and space period that gives us better representationof different meteorological conditions, dust transport paths,500

and measurement locations. However, the considerations andmetrics presented here can also be applied to the NRT evalu-ation scenario.

As was shown in section 2.2, synergistic retrieval algo-rithms can help avoiding possible comparison biases caused505

by the presence of aerosol mixtures, by separating the dustcontribution on the total aerosol measurement. Even whenusing these algorithms, however, the direct comparison withdust models should be done carefully, because the quantitiesidentifies as dust in the measurements could differ. For ex-510

ample, in the case of the POLIPHON algorithm (Ansmannet al., 2012), dust is identified as the only depolarizing com-ponent of measured lidar signals. In the case of LIRIC, dustis assumed to be a particle component larger than ∼ 0.5µmin radius. On the other hand, the total dust load predicted by515

the models includes also smaller particle sizes in the first fewbins of the dust size distribution. The contribution of thesesmall particles in the total aerosol load should be typicallylow, especially near the source (d’Almeida, 1987), but couldbecome more important in cases of long-range dust trans-520

port where the larger particles are gravitationally removed(Mamouri and Ansmann, 2014). When using a statistical ap-proach including different location and transport paths, as inthe present study, this effect is expected to be negligible, butcould become important when performing a case study eval-525

uation; in the latter case only specific bins from the modeloutput should be used instead.

In the case of statistical model evaluation, the selectedmeasurement profiles should also be independent to give acorrect representation of the model’s performance. Specifi-530

cally, it should be avoided that the used measurements fromeach station sample the same event multiple times, but shouldinstead measure independent dust transport events. This con-sideration is less important when using data from automaticinstruments; in the case of EARLINET, however, the avail-535

able dataset could be affected by data from long observationsperiods and intensive measurement campaigns, as describedin section 2.1. Ideally, only a climatological dataset would beused, but the number of the available case would be limitedfrom the measurement frequency, the sporadic nature of dust540

transport episodes, and, when using synergistic algorithms,the availability of AERONET data. In this study we considerindependent measurement that had at least 24 hour time dif-ference, compatible with the expected variability of tropo-spheric aerosols (Anderson et al., 2003a, b).545

The vertical resolution of lidar and dust model profilesshould be taken into account during their comparison. Thelidar signals have a raw vertical resolution of few meters andthe final products have an effective resolution of few hundredmeters depending on filtering procedures and smoothness550

constraints used in the retrieval (Pappalardo et al., 2004). Thevertical resolution of the models, on the other hand, is typi-cally coarser but depends on the vertical resolution of the me-


teorological driver (Simmons and Burridge, 1981; Mesinger,1984). When performing a statistical comparison, the differ-555

ent vertical resolutions are less important as the features ofindividual dust transport cases will be smoothed. When com-paring aerosol extensive properties (both optical and concen-trations) the remote sensing profiles should be upscaled to themodel resolution. When, however, the aim of the comparison560

is to evaluate the dust layer geometrical properties and valuesat specific location, e.g. the peak values, the full resolutionremote sensing profiles should be used, as the model’s verti-cal resolution is the parameter that is evaluated. In this study,to facilitate the comparison of models of different vertical565

resolution, we interpolate all available profiles to a common100m vertical resolution. We used this resolution to performthe geometrical property and peak value evaluation but used500m averages to calculate the statistics on the vertical pro-files presented in the next section. The models simulate the570

dust concentration profiles on a specified horizontal grid, sobilinear interpolation was used to estimate the concentrationvalues at the exact location for each station. Linear interpo-lation was also utilized to estimate the concentration profilesat the exact time of the available measurements.575

Correct representation of the dust mixing in the PBL canimpact in the forecasted air quality and also affect the re-moval processes of dust in the model. In this process, dustis mixed with locally produced aerosols, so lidar optical pro-files cannot be used to study the dust effect. The mass re-580

trieval algorithms, like LIRIC, can separate the dust com-ponent in the PBL and give some insights to study this pro-cess, even though several limitation remain. Firstly, local dustsources could contribute to the dust load in the PBL (Korczet al., 2009) but the exact effect of such sources to the ver-585

tical dust distribution, in our knowledge, has not been sys-tematically studied. Secondly, as dust comes in contact withother types of particles and high relative humidity, some ofthe assumptions of the retrieval algorithms could be invalid.For example, dust can be coated and water layer can form on590

the dust particles, changing their optical properties like thedepolarization ratio. Such effect could be important for theexact quantitative characterization of dust but still allow thestudy of the mixing of dust in the PBL. Lastly, most lidarsystems have a high overlap function and can only study the595

initial mixing of dust in the upper parts of the PBL. Giventhese factors, the study of this mixing process could be donebetter for specific case studies. If a statistical approach is fol-lowed, the dataset should be large enough to give significantresults, as only few profiles cannot capture this dynamical600

mixing phenomenon. For example, in the dataset presentedin the next section, possible mixing of dust in the PBL wasobserver only in few cases, thus the lower part of the profilewas excluded.

The direct output of all the above mentioned algorithms is605

volume concentration profiles of fixed aerosol types. This canbe converted to mass concentration profiles, the typical out-put of dust transport models, by using the aerosol bulk den-

sity. In the case of dust, the typically used value is 2.6g/cm3

(Köpke et al., 1997; Ansmann et al., 2012)) while the actual610

bulk concentration could differ by location (e.g. Todd et al.(2007)). In the case of dust model evaluation, however, se-lecting a value of 2.6g/cm3 is compatible with the assump-tions of most dust transport models (Tegen and Fung, 1994;Nickovic et al., 2001; Yumimoto et al., 2012), and thus a fur-615

ther reason for discrepancies is removed from this study.We perform the comparison firstly by examining single pa-

rameters of each measurement case and secondly looking inprofile properties to capture both the total performance of themodels and the detailed performance per altitude. The single620

parameters examined are total concentration, peak value ofthe profile, and center of mass of the predicted dust verticaldistribution. For the profile parameters, apart from the aver-age profiles, we examine the mean bias error, correlation co-efficient, root mean square error, and fractional gross error.625

This set of parameters was chosen because they can providea detailed view of performance and trying to remain compat-ible, as much as possible, with the metrics already in use inthe SDS-WAS columnar evaluation.

The first single-value measure to compare is the dust to-630

tal concentration, calculated across the altitude range whereboth measured and model profiles provide valid results. Inthis way, this comparison will be a little different than com-paring directly columnar measurements, as in the case ofcomparing photometer and total column model values. In the635

latter case the used range excludes the lower few hundredmeters of the profile, thus including the contribution of localdust sources to the total column aerosol load, possibly pro-ducing a bias in the measurements.

A second metric examined is the peak value of the profile,640

a value typically indicating the main vertical location of thedust plume. In cases where the main dust mass is locatednear the ground, the lidar system can fail to detect the truemaximum, and instead show a maximum value at the lowestpoint of the profile, i.e. first point of full overlap. In these645

cases we considered as maximum value the first lofted layerpeak, located as the first peak after the first local minimumof the concentration profile.

An important indicator for model vertical profiles is thecenter of mass (CoM), a parameter that gives in a single num-650

ber an indication of the altitude of the dust distribution. Incases were a single aerosol layer is present in the atmosphere,the CoM gives an indication of its mean altitude; in case ofmultiple layers, however, the CoM could be located in areaswithout any considerable dust load (Mona et al., 2006, 2014).655

The profile properties were calculated by first averagingthe compared profiles at 500m resolution and a set of statis-tics were calculated for each altitude range. This resolutionwas chosen as a compromise between the need to distin-guish detailed aerosol structure and the signal noise of the660

lidar measurements. This value, however, needs to be deter-mined in each study based on the number of available pro-files measurements. Apart of the mean value profiles, the


first set of measures used are the mean bias, and the rootmean square error (RMSE); being expressed in units of con-665

centration, these values are suitable for the intercompari-son of models but can be misleading for the performance ofmodels with altitude. In addition, RMSE is strongly domi-nated by the largest values, due to the squaring operation andin cases where prominent outliers occur, the usefulness of670

RMSE is questionable and the interpretation becomes moredifficult. These limitations are addressed using a second setof measurements, including correlation coefficient, fractionalbias, and fractional gross error. These measures don’t dependon the absolute value of the observation, are bounded, and675

handle symmetrically the effect of model under- and over-estimation. Definitions of the used measures are given in Ta-ble 3.

4 Application

In this section we demonstrate the application of the de-680

scribed dust model evaluation methodology to simulationsperformed by the four models described in section 2.3. TenEuropean remote sensing stations contributed data in thisintercomparison, mainly concentrated in the Mediterraneanarea, as shown in Figure 1; their details can be seen Table 4.685

All stations are part of the EARLINET and AERONET net-works, a fact that guarantees that the provided data are of uni-form quality. The participating stations collected, in total, 69dust profiles of dust measurements. The cases span the periodfrom February 2011 to June 2013, with the majority located690

in spring and summer period (see Figure 2)), when most Sa-haran dust transport episodes occur and clouds free condi-tions, needed for the measurements, are usually found (Monaet al., 2006; Papayannis et al., 2008). The stations selectedthe cases independently, based on the available measurement.695

For each station, the selected profiles were screened for hav-ing at least 24h time distance, as described before, to consideronly measurements of different dust transport events. The ac-tual number of available measurements varies with altitude asshown in Fig. 3. In the lower altitudes, the number is limited700

by the ground level altitude of the stations and the incompletemeasurement range of the instruments. In the higher altitudethe lidar profiles were cut at the points were dust was de-tected for each measurement case. The four examined dusttransport models were run for the given period and the out-705

put was stored for three hour intervals. For each time step,bilinear interpolation was used to estimate the concentrationvalues at the exact location for each station. Linear interpo-lation was also used to estimate the concentration profiles atthe exact time of the available measurements.710

The comparison based on center of mass reveals thatmodels can correctly track the main vertical location oftransport dust. In Fig. 4 this comparison is presented forthe four models, and shows that the models predict thedust CoM in almost all cases. The BSC/DREAM8bV2 and715

DREAMABOL models show almost zero bias tracking thelocation of dust almost perfectly, except in few outlyingcases. Instead, NMMB/BSC-DUST and DREAM8-NMME-MACC overestimate the center of mass altitude, especiallyin cases with observed CoM above 3km; the fractional720

bias values for NMMB/BSC-DUST and DREAM8-NMME-MACC are 0.17 and 0.13 respectively. The correlation co-efficient, especially for BSC/DREAM8bV2, is determinedby few extreme cases; the values for the four models are0.38 for BSC/DREAM8bV2, 0.54 for DREAMABOL, 0.58725

for NMMB/BSC-DUST and 0.64 for DREAM8-NMME-MACC.

Our examination indicates that the simulations systemati-cally underestimate the total amount of dust. Fig. 5 presentsthe comparison of the dust concentration integrated across730

the common altitude range. The volume concentration fromthe four models shows significant correlation with the mea-sured one, but in general is underestimated. For high concen-tration cases (values greater than ∼ 0.3g/m2), NMMB/BSC-DUST and DREAM8-NMME-MACC predictsufficiently the735

concentration values, while the other two models tend to un-derestimate. In low concentration values (less than 0.3g/m2)all models apart from DREAM8-NMME-MACC, underes-timate in many cases the dust concentration. This could becaused by insufficient dust source strength, overestimated de-740

position and wet scavenging parameters, or a combinationof both; however the current dataset is not sufficient to dis-crimintae the exact factor affection the comparison from themodel point of view. However, it is believed that using thepresent approach as part of a complete, multi-sensor evalua-745

tion exercise help evaluating possible model limitations. Thetotal fractional bias values for the models range from -1.04 to-0.22, while correlation coefficients range from 0.24 to 0.64.

Fig. 6 shows the relationship of peak simulated values foreach profile and the measured ones. Also in this case, the750

models underestimate the maximum value of each profile.The fractional bias for the four models ranges from -0.90 to-0.31 while the correlation coefficient ranges has smaller val-ues than before from 0.19 to 0.48. This result can only partlybe explained by the overall concentration underestimation755

that was noted before. The lower resolution of the models,especially in higher altitudes, could lead to a "smoothing"effect of individual peak values in the compared cases. Asimilar effect could be caused by the mixing of the dust inall the volume of the model’s grid. A summary of the above760

described statistics for all the examined models is given inTable 5.

In summary, the current study indicates that the examineddust models represent well the altitude of transport whilethe total concentration is predicted lower than measured,765

with sharp peaks smoothed out. The performance of modelsin specific cases, and also between them, can vary signifi-cantly. Figure 7 summariezes the performance of all modelsin a case-by-case comparison. For each model-measurementpair we calculate the correlation coefficient with altitude as770


well as the fractional bias, and the results are plotted ina scatterplot. The ideal model would have correlation one,i.e. it would predict perfectly the shape of the dust profile,and ∼ 0 fractional bias, i.e. predicting correctly the quan-tity of transported dust. While individual cases show a big775

variability, each model shows a characteristic pattern. ForBSC/DREAM8bV2 and DREAMABOL most cases havehigh correlation but negative fractional bias i.e. the modelscan often predict correctly the shape of the dust profile butunderestimate the concentration. In contrast, NMMB/BSC-780

DUST and DREAM8-NMME-MACC have fractional biasvalue distribution near 0 but a wider spread of correlationvalues. For all models, it should be observed that there is aconsiderable spread of values for the specific comparisons, afurther argument for the need for a statistical evaluation of785

dust model’s performance.These results are further supported by directly comparing

the profile data provided by the model, indicating the mod-els don’t only capture the general altitude of dust transportbut, in average, predict correctly the shape of the dust pro-790

file. In Fig. 8 the mean measured profile for all 69 cases iscompared with the corresponding profiles of the four mod-els. The profiles show good agreement in the predicted shapeof the dust concentration, but have wider spread in the abso-lute values. BSC/DREAM8bV2 and DREAMABOL predicts795

a maximum dust concentration in the region 2 - 3 km a.s.l.,in agreement with the observation while the other two mod-els have the maximum value at slightly higher altitude 3 - 4km. DREAM8-NMME-MACC overestimated the concentra-tion of dust in altitudes above ∼ 5km; specifically, while the800

observed values of dust are below 10µg/m3 above 6km, themodel predicts these values only above 8km. The concentra-tion values show wider discrepancy: while the peak value ofthe mean profiles is retrieved ∼ 65µg/m3 the models peakvalues range from ∼ 30µg/m3 to ∼ 50µg/m3. When ex-805

amining the profile data, we can observe the differences inhigh and low concentration cases that were described before,as shown in Figure 9; NMMB/BSC-DUST and DREAM8-NMME-MACC have particularly good agreement at the highconcentration cases. As noted before, such finding highlight810

the importance of statistical evaluation studies and indicatethat this trend should be investigated in a future completeevaluation study.

The above results are further explored in figures 10 to13. Figure 10 presents the mean bias of the four studied815

models. All models show negative bias below 4km whileabove that altitude NMMB/BSC-DUST has almost 0 biasand DREAM8-NMME-MACC positive bias values. At thealtitude range where most dust is located, i.e. from 2 to 4 kma.s.l., the bias ranges from −40µg/m3 to 0µg/m3. In figure820

11 the variation of the RMSE with altitude is shown; in the2 - 4 km range the mean values range from 45 to 55µg/m3

with the maximum value ∼ 70µg/m3 reached by DREAM-ABOL at 2km a.s.l. The profiles of correlation coefficientfor the four models is shown in figure 12. All four models825

show significant correlation for altitude range from 1 to 6 km,which is the region which practically all dust is typically ob-served. The mean values range from 0.5 for DREAMABOLto 0.7 for DREAM8-NMME-MACC. Finally, in figure 13 thefractional gross error is shown. The minimum values for FE ,830

ranging from 0.7 to 1.25, are observed at 2-3km. In higheraltitudes, the FE values are higher, with values ranging from1.25 to 1.6 at 6km a.s.l.

A summary of the different behavior of the four modelsis given in Figure 14 using Taylor diagrams (Taylor, 2001).835

The data of the models and measurements were averaged at1km altitude range before calculating the statistics. Four Tay-lor diagrams are presented, for the altitude range from 1 to5 km. DREAM8-NMME-MACC seems to capture correctlythe variability of dust event in all altitude ranges, a property840

that can be partly attribute to the use of data assimilation.NMMB/BSC-DUST show similar good performance, espe-cially for 3 to 5 km. As it was observer before, the other twomodels underestimate the variability of dust in a consistentway with altitude. The model simulations show good corre-845

lation at all four altitude ranges.The presented results depend on regional and seasonal

variations. While the available cases in this study are not suf-ficient to perform a seasonal analysis, they can be used to geta hint of the insight that can be gained from a regional eval-850

uation of the model’s performance. With this aim, the avail-able stations were divided in two clusters, a west and an eastone. The west cluster of stations, including Evora, Granada,and Barcelona, is affected by dust events arriving only a fewdays of transport. The east cluster, including Potenza, Lecce,855

Athens, Thessaloníki, and Bucharest, is affected by longertransport of dust from both the West and Central Sahara. Theregional comparison of the mean dust concentration profilesis shown in figure 15. The average profiles indicate that thedust is transported in different altitudes, with maximum value860

observed around 2km a.s.l. for the west cluster and around3km a.s.l. for the east cluster, a behavior that is well cap-tured by all models. The correlation coefficient at all altitudesis higher for the East rather than the west cluster as shownin Figure 16. Specifically, the average correlation at altitude865

range from 2 to 6 km ranges for the west cluster from 0.46to 0.62 and from 0.55 to 0.83 for the east cluster. This differ-ence can be attributed to the strong effect of orography on thewest cluster, as Atlas Mountains and orography of the IberianPeninsula make the prediction of the dust transport difficult,870

while the transport to the East cluster is performed, for largepart of the transport path, over the Mediterranean Sea. Addi-tionally, the longer transport to the east cluster homogenizesthe dust transport event and makes small inconsistencies inspace and time less relevant. These preliminary results indi-875

cate that the regional aspects in prediction of vertical distri-bution of dust should be further studied.


5 Conclusions

We presented here an approach for using synergy ofcontinental-scale remote sensing networks for the evalua-880

tion of dust transport models. The development and use ofLIRIC retrieval algorithm within ACTRIS/EARLINET per-mits the estimation of dust volume concentration profiles,making possible the direct comparison with dust models. Thecoordinated lidar and photometer measurements that are sys-885

tematically performed across Europe allows the evaluation ofthe distribution of dust under different meteorological condi-tions, locations, and transport paths. These factors combinedconstitute an important new tool for the evaluation of dustmodel forecasts over Europe.890

We presented a methodology for the construction of theappropriate evaluation dataset for the dust model evaluationstudy, considering the aspects of the model that need to bestudied and the potential and limitations of the remote sens-ing data. A set of metrics, compatible with the current SDS-895

WAS evaluation procedures, were also presented, adapted tothe study of vertical component of model’s output. As an ex-ample, the above methodology and metrics were applied forthe examination of four regional dust transport models thatcontribute to SDS-WAS. This initial comparison revealed the900

similarities and differences in the performance of these trans-port models, and highlighted the benefit that could be gainedby systematic use of the proposed approach in their evalua-tion.

Sophisticated algorithms described in this paper — such905

as LIRIC, GARRLiC, POLIPHON — developed within AC-TRIS, integrate EARLINET lidar and AERONET sun pho-tometer observations for an improved characterization of thefour-dimensional distribution of aerosols over Europe. Theseobservation can be used, in coordination with other evalua-910

tion methodologies, for the improvement of dust model per-formance and lead to better understanding the effect of dustin air-quality, weather, and the climate.

Acknowledgements. The financial support of the ACTRIS ResearchInfrastructure Project supported by the European Union Seventh915

Framework Programme (FP7/2007-2013) under grant agreementn. 262254 is gratefully acknowledged. This project has also re-ceived funding from the European Union’s Seventh Framework Pro-gramme for research, technological development and demonstrationunder grant agreement no 289923 – ITaRS.920

References

Anderson, T. L., Charlson, R. J., Winker, D. M., Ogren, J. A., andHolmén, K.: Mesoscale Variations of Tropospheric Aerosols, J.Atmos. Sci., 60, 119–136, 2003a.

Anderson, T. L., Masonis, S. J., Covert, D. S., Ahlquist, N. C.,925

Howell, S. G., Clarke, A. D., and McNaughton, C. S.: Vari-ability of aerosol optical properties derived from in situ aircraft

measurements during ACE-Asia, J. Geophys. Res., 108, 8647,doi:10.1029/2002JD003247, 2003b.

Ansmann, A., Tesche, M., Seifert, P., Groß, S., Freudenthaler, V.,930

Apituley, A., Wilson, K. M., Serikov, I., Linné, H., Heinold, B.,Hiebsch, A., Schnell, F., Schmidt, J., Mattis, I., Wandinger, U.,and Wiegner, M.: Ash and fine-mode particle mass profiles fromEARLINET-AERONET observations over central Europe afterthe eruptions of the Eyjafjallajökull volcano in 2010, J. Geophys.935

Res., 116, D00U02, doi:10.1029/2010JD015567, 2011.Ansmann, A., Seifert, P., Tesche, M., and Wandinger, U.: Profiling

of fine and coarse particle mass: case studies of Saharan dustand Eyjafjallajökull/Grimsvötn volcanic plumes, Atmos. Chem.Phys., 12, 9399–9415, 2012.940

Barbosa, H. M. J., Lopes, F. J. S., Silva, A., Nisperuza, D., Barja,B., Ristori, P., Gouveia, D. A., Jimenez, C., Montilla, E., Mari-ano, G. L., and others: The first ALINE measurements and inter-comparison exercise on lidar inversion algorithms, Optica pura yaplicada, 47, 99–108, 2014.945

Basart, S., Pérez, C., Nickovic, S., Cuevas, E., and Baldasano,J. M.: Development and evaluation of the BSC-DREAM8b dustregional model over Northern Africa, the Mediterranean andthe Middle East, Tellus B, 64, doi:10.3402/tellusb.v64i0.18539,2012.950

Benedetti, A., Morcrette, J.-J., Boucher, O., Dethof, A., Engelen,R. J., Fisher, M., Flentje, H., Huneeus, N., Jones, L., Kaiser,J. W., Kinne, S., Mangold, A., Razinger, M., Simmons, A. J.,and Suttie, M.: Aerosol analysis and forecast in the EuropeanCentre for Medium-Range Weather Forecasts Integrated Forecast955

System: 2. Data assimilation, J. Geophys. Res., 114, D13 205,doi:10.1029/2008JD011115, 2009.

Benedetti, A., Baldasano, J. M., Basart, S., Benincasa, F., Boucher,O., Brooks, M. E., Chen, J.-P., Colarco, P. R., Gong, S., Huneeus,N., Jones, L., Lu, S., Menut, L., Morcrette, J.-J., Mulcahy, J.,960

Nickovic, S., García-Pando, C. P., Reid, J. S., Sekiyama, T. T.,Tanaka, T. Y., Terradellas, E., Westphal, D. L., Zhang, X.-Y., and Zhou, C.-H.: Operational Dust Prediction, in: MineralDust, edited by Knippertz, P. and Stuut, J.-B. W., pp. 223–265,Springer Netherlands, 2014.965

Betts, A. K.: A new convective adjustment scheme. Part I: Observa-tional and theoretical basis, Q.J.R. Meteorol. Soc., 112, 677–691,doi:10.1002/qj.49711247307, 1986.

Buzzi, A., D’Isidoro, M., and Davolio, S.: A case-study of an oro-graphic cyclone south of the Alps during the MAP SOP, Q.J.R.970

Meteorol. Soc., 129, 1795–1818, doi:10.1256/qj.02.112, 2003.Böckmann, C., Wandinger, U., Ansmann, A., Bösenberg, J.,

Amiridis, V., Boselli, A., Delaval, A., De Tomasi, F., Frioud, M.,Grigorov, I. V., Hågård, A., Horvat, M., Iarlori, M., Komguem,L., Kreipl, S., Larchevêque, G., Matthias, V., Papayannis, A.,975

Pappalardo, G., Rocadenbosch, F., Rodrigues, J. A., Schneider,J., Shcherbakov, V., and Wiegner, M.: Aerosol lidar intercom-parison in the framework of the EARLINET project. 2. Aerosolbackscatter algorithms, Appl. Opt., 43, 977–989, 2004.

Bösenberg, J., Ansmann, A., Baldasano, J. M., Calpini, B.,980

Chaikovsky, A., Flamant, P., Mitev, V., Flamant, A., Hågård, A.,Mitev, V., Papayannis, A., Pelon, J., Resendes, D., Schneider, J.,Spinelli, N., Trickl, T., Vaughan, G., Visconti, G., and Wiegner,M.: EARLINET : a European aerosol research lidar network, in:Advances in laser remote sensing, edited by Dabas, A., Loth, C.,985

and Pelon, J., Ecole polytechnique, 2001.


Chaikovsky, A.: Lidar-Radiometer Inversion Code (LIRIC) forthe retrieval of vertical aerosol properties using combinedlidar/radiometer data, Atmospheric Measurement Techniques,2015.990

Chaikovsky, A., Dubovik, O., Goloub, P., Tanré, D., Pappalardo,G., Wandinger, U., Chaikovskaja, L., Denisov, S., Grudo, Y.,Lopatsin, A., Karol, Y., Lapyonok, T., Korol, M., Osipenko, F.,Savitski, D., Slesar, A., Apituley, A., Alados-Arboledas, L., Bini-etoglou, I., Kokkalis, P., Munoz, M. J. G., Papayannis, A., Per-995

rone, M. R., Pietruczuk, A., Pisani, G., Rocadenbosch, F., Sicard,M., De Tomasi, F., and Wang, X.: Algorithm and software forthe retrieval of vertical aerosol properties using combined li-dar/radiometer data: dissemination in EARLINET, in: Reviewedand Revised Papers of the 26th ILRC, edited by Papayannis,1000

A., Balis, D., and Amiridis, V., vol. 1, pp. 399–402, Porto Heli,Greece, 2012.

Cuesta, J., Flamant, P. H., and Flamant, C.: Synergetic tech-nique combining elastic backscatter lidar data and sunphotome-ter AERONET inversion for retrieval by layer of aerosol opti-1005

cal and microphysical properties, Appl. Opt., 47, 4598–4611,doi:10.1364/AO.47.004598, 2008.

d’Almeida, G. A.: On the variability of desert aerosol ra-diative characteristics, J. Geophys. Res., 92, 3017–3026,doi:10.1029/JD092iD03p03017, 1987.1010

Dubovik, O.: Optimization of numerical inversion in photopolari-metric remote sensing, in: Photopolarimetry in remote sensing,pp. 65–106, Springer, 2005.

Dubovik, O. and King, M. D.: A flexible inversion algorithm forretrieval of aerosol optical properties from Sun and sky radiance1015

measurements, J. Geophys. Res., 105, 673–20, 2000.Dubovik, O., Sinyuk, A., Lapyonok, T., Holben, B. N., Mishchenko,

M., Yang, P., Eck, T. F., Volten, H., Muñoz, O., Veihelmann, B.,van der Zande, W. J., Leon, J.-F., Sorokin, M., and Slutsker, I.:Application of spheroid models to account for aerosol particle1020

nonsphericity in remote sensing of desert dust, J. Geophys. Res.,111, D11 208, doi:10.1029/2005JD006619, 2006.

Ferrier, B. S., Jin, Y., Lin, Y., Black, T., Rogers, E., and DiMego,G.: Implementation of a new grid-scale cloud and precipitationscheme in the NCEP Eta model, in: Conference on weather anal-1025

ysis and forecasting, vol. 19, pp. 280–283, AMS, 2002.Ginoux, P., Chin, M., Tegen, I., Prospero, J. M., Holben, B.,

Dubovik, O., and Lin, S.-J.: Sources and distributions of dustaerosols simulated with the GOCART model, J. Geophys. Res.Atmos., 106, 20 255–20 273, doi:10.1029/2000JD000053, 2001.1030

Gobbi, G. P., Angelini, F., Barnaba, F., Costabile, F., Baldasano,J. M., Basart, S., Sozzi, R., and Bolignano, A.: Changes in par-ticulate matter physical properties during Saharan advectionsover Rome (Italy): a four-year study, 2001–2004, Atmos. Chem.Phys., 13, 7395–7404, doi:10.5194/acp-13-7395-2013, 2013.1035

Granados-Muñoz, M. J., Guerrero-Rascado, J. L., Bravo-Aranda,J. A., Navas-Guzmán, F., Valenzuela, A., Lyamani, H.,Chaikovsky, A., Wandinger, U., Ansmann, A., Dubovik, O.,Grudo, J. O., and Alados-Arboledas, L.: Retrieving aerosolmicrophysical properties by Lidar-Radiometer Inversion Code1040

(LIRIC) for different aerosol types, J. Geophys. Res. Atmos.,119, 2013JD021 116, 2014.

Guerrero-Rascado, J. L., Olmo, F. J., Avilés-Rodríguez, I., Navas-Guzmán, F., Pérez-Ramírez, D., Lyamani, H., and Alados Ar-boledas, L.: Extreme Saharan dust event over the southern1045

Iberian Peninsula in september 2007: active and passive remotesensing from surface and satellite, Atmos. Chem. Phys., 9, 8453–8469, 2009.

Haustein, K., Pérez, C., Baldasano, J. M., Jorba, O., Basart, S.,Miller, R. L., Janjic, Z., Black, T., Nickovic, S., Todd, M. C.,1050

Washington, R., Müller, D., Tesche, M., Weinzierl, B., Esselborn,M., and Schladitz, A.: Atmospheric dust modeling from mesoto global scales with the online NMMB/BSC-Dust model – Part2: Experimental campaigns in Northern Africa, Atmos. Chem.Phys., 12, 2933–2958, 2012.1055

Hess, M., Koepke, P., and Schult, I.: Optical properties of aerosolsand clouds: The software package OPAC, B. Am. Meteorol. Soc.,79, 831–844, 1998.

Holben, B. N., Eck, T. F., Slutsker, I., Tanre, D., Buis, J. P., Set-zer, A., Vermote, E., Reagan, J. A., Kaufman, Y. J., and Naka-1060

jima, T.: AERONET—A federated instrument network and dataarchive for aerosol characterization, Remote sens. environ., 66,1–16, 1998.

Janjic, Z., Janjic, T., and Vasic, R.: A Class of Conservative Fourth-Order Advection Schemes and Impact of Enhanced Formal Ac-1065

curacy on Extended-Range Forecasts, Mon. Wea. Rev., 139,1556–1568, doi:10.1175/2010MWR3448.1, 2011.

Janjic, Z. I., Gerrity, J. P., and Nickovic, S.: An Al-ternative Approach to Nonhydrostatic Modeling,Mon. Wea. Rev., 129, 1164–1178, doi:10.1175/1520-1070

0493(2001)129<1164:AAATNM>2.0.CO;2, 2001.Janjic, Z. I.: The Step-Mountain Eta Coordinate Model: Further

Developments of the Convection, Viscous Sublayer, and Tur-bulence Closure Schemes, Mon. Wea. Rev., 122, 927–945,doi:10.1175/1520-0493(1994)122<0927:TSMECM>2.0.CO;2,1075

1994.Jiménez-Guerrero, P., Pérez, C., Jorba, O., and Baldasano, J. M.:

Contribution of Saharan dust in an integrated air quality sys-tem and its on-line assessment, Geophys. Res. Lett., 35, L03 814,doi:10.1029/2007GL031580, 2008.1080

Kim, S.-W., Yoon, S.-C., Dutton, E. G., Kim, J., Wehrli, C., andHolben, B. N.: Global Surface-Based Sun Photometer Networkfor Long-Term Observations of Column Aerosol Optical Prop-erties: Intercomparison of Aerosol Optical Depth, Aerosol sci.tech., 42, 1–9, doi:10.1080/02786820701699743, 2008.1085

Kokkalis, P., Papayannis, A., Mamouri, E, R., Tsaknakis, G., andAmiridis, V.: The EOLE lidar system of the National Techni-cal University of Athens, in: Reviewed and revised papers pre-sented at the 26th ILRC, edited by Papayannis, A., Balis, D., andAmiridis, V., pp. 629–632, Porto Heli, Greece, 2012.1090

Korcz, M., Fudała, J., and Klis, C.: Estimation of wind blown dustemissions in Europe and its vicinity, Atmos. Environ., 43, 1410–1420, doi:10.1016/j.atmosenv.2008.05.027, 2009.

Kumar, D., Rocadenbosch, F., Sicard, M., Comeron, A., Muñoz, C.,Lange, D., Tomás, S., and Gregorio, E.: Six-channel polychro-1095

mator design and implementation for the UPC elastic/Raman LI-DAR, in: SPIE Remote Sensing, pp. 81 820W–81 820W, Interna-tional Society for Optics and Photonics, 2011.

Köpke, P., Hess, M., Schult, I., and Shettle, E. P.: Global aerosoldata set, Max-Planck-Institut für Meteorologie Hamburg, Ger-1100

many, 1997.Lewandowski, P. A., Eichinger, W. E., Holder, H., Prueger, J.,

Wang, J., and Kleinman, L. I.: Vertical distribution of aerosols inthe vicinity of Mexico City during MILAGRO-2006 Campaign,


Atmos. Chem. Phys., 10, 1017–1030, doi:10.5194/acp-10-1017-1105

2010, 2010.Lopatin, A., Dubovik, O., Chaikovsky, A., Goloub, P., Lapyonok,

T., Tanré, D., and Litvinov, P.: Enhancement of aerosol charac-terization using synergy of lidar and sun-photometer coincidentobservations: the GARRLiC algorithm, Atmos. Meas. Tech., 6,1110

2065–2088, doi:10.5194/amt-6-2065-2013, 2013.Madonna, F., Amodeo, A., Boselli, A., Cornacchia, C., Cuomo, V.,

D’Amico, G., Giunta, A., Mona, L., and Pappalardo, G.: CIAO :the CNR-IMAA advanced observatory for atmospheric research,Atmos. Meas. Tech., 4, 1191–1208, 2011.1115

Mamouri, R. E. and Ansmann, A.: Fine and coarse dust separa-tion with polarization lidar, Atmos. Meas. Tech., 7, 3717–3735,doi:10.5194/amt-7-3717-2014, 2014.

Marticorena, B. and Bergametti, G.: Modeling the atmospheric dustcycle: 1. Design of a soil-derived dust emission scheme, J. Geo-1120

phys. Res., 100, 16 415–16 430, doi:10.1029/95JD00690, 1995.Matthias, V., Freudenthaler, V., Amodeo, A., Balin, I., Balis, D.,

Bösenberg, J., Chaikovsky, A., Chourdakis, G., Comeron, A.,Delaval, A., De Tomasi, F., Eixmann, R., Hågård, A., Komguem,L., Kreipl, S., Matthey, R., Rizi, V., Rodrigues, J. A., Wandinger,1125

U., and Wang, X.: Aerosol lidar intercomparison in the frame-work of the EARLINET project. 1. Instruments, Appl. Opt., 43,961–976, 2004.

Maurizi, A. and Monti, F.: DREAMABOL 1.0 – A BOLCHEM off-spring for mineral dust. In preparation for Geoscientific Model1130

Development, Geoscientific Model Development Discussions.Maurizi, A., D’Isidoro, M., and Mircea, M.: BOLCHEM: An In-

tegrated System for Atmospheric Dynamics and Composition,in: Integrated Systems of Meso-Meteorological and ChemicalTransport Models, pp. 89–94, Springer, 2011.1135

Mesinger, F.: A blocking technique for representation of mountainsin atmospheric models, Riv. Meeteorol. Aeronau., 44, 195–202,1984.

Mircea, M., D’Isidoro, M., Maurizi, A., Vitali, L., Monforti, F.,Zanini, G., and Tampieri, F.: A comprehensive performance eval-1140

uation of the air quality model BOLCHEM to reproduce theozone concentrations over Italy, Atmospheric Environment, 42,1169–1185, doi:10.1016/j.atmosenv.2007.10.043, 2008.

Mona, L., Amodeo, A., Pandolfi, M., and Pappalardo, G.: Saha-ran dust intrusions in the Mediterranean area: Three years of1145

Raman lidar measurements, J. Geophys. Res., 111, D16 203,doi:10.1029/2005JD006569, 2006.

Mona, L., Papagiannopoulos, N., Basart, S., Baldasano, J., Bini-etoglou, I., Cornacchia, C., and Pappalardo, G.: EARLINETdust observations vs. BSC-DREAM8b modeled profiles: 12-1150

year-long systematic comparison at Potenza, Italy, Atmos. Chem.Phys., 14, 8781–8793, doi:10.5194/acp-14-8781-2014, 2014.

Nemuc, A., Vasilescu, J., Talianu, C., Belegante, L., and Nicolae,D.: Assessment of aerosol’s mass concentrations from measuredlinear particle depolarization ratio (vertically resolved) and sim-1155

ulations, Atmos. Meas. Tech., 6, 3243–3255, 2013.Nickovic, S., Kallos, G., Papadopoulos, A., and Kakaliagou, O.: A

model for prediction of desert dust cycle in the atmosphere, J.Geophys. Res., 106, 18 113–18, 2001.

Nickovic, S., Vukovic, A., and Vujadinovic, M.: Atmospheric pro-1160

cessing of iron carried by mineral dust, Atmos. Chem. Phys., 13,9169–9181, 2013.

Nickovic, S. and Dobricic, S.: A model for long-range transport ofdesert dust, Mon. weather rev., 124, 2537–2544, 1996.

Papayannis, A., Amiridis, V., Mona, L., Tsaknakis, G., Balis, D.,1165

Bösenberg, J., Chaikovski, A., De Tomasi, F., Grigorov, I., Mat-tis, I., Mitev, V., Müller, D., Nickovic, S., Pérez, C., Pietruczuk,A., Pisani, G., Ravetta, F., Rizi, V., Sicard, M., Trickl, T., Wieg-ner, M., Gerding, M., Mamouri, R. E., D’Amico, G., and Pap-palardo, G.: Systematic lidar observations of Saharan dust over1170

Europe in the frame of EARLINET (2000–2002), J. Geophys.Res., 113, 2008.

Papayannis, A., Balis, D., Kokkalis, P., Mamouri, R., Tsaknakis,G., Giannakaki, E., Siomos, N., and Amiridis, V.: ARIADNE:The Greek lidar network, in: Reviewed and Revised Papers of the1175

26th ILRC, edited by Papayannis, A., Balis, D., and Amiridis, V.,pp. 349–352, Porto Heli, Greece, 2012.

Pappalardo, G., Amodeo, A., Pandolfi, M., Wandinger, U., Ans-mann, A., Bösenberg, J., Matthias, V., Amiridis, V., De Tomasi,F., Frioud, M., Iarlori, M., Komguem, L., Papayannis, A., Roca-1180

denbosch, F., and Wang, X.: Aerosol lidar intercomparison in theframework of the EARLINET project. 3. Raman lidar algorithmfor aerosol extinction, backscatter, and lidar ratio, Appl. Opt., 43,5370–5385, 2004.

Pappalardo, G., Amodeo, A., Apituley, A., Comeron, A., Freuden-1185

thaler, V., Linné, H., Ansmann, A., Bösenberg, J., D’Amico,G., Mattis, I., Mona, L., Wandinger, U., Amiridis, V., Alados-Arboledas, L., Nicolae, D., and Wiegner, M.: EARLINET: to-wards an advanced sustainable European aerosol lidar network,Atmos. Meas. Tech., 7, 2389–2409, 2014.1190

Perrone, M. R., De Tomasi, F., and Gobbi, G. P.: Vertically resolvedaerosol properties by multi-wavelength lidar measurements, At-mos. Chem. Phys., 14, 1185–1204, 2014.

Pietruczuk, A. and Chaikovsky, A.: Variability of aerosol proper-ties during the 2007–2010 spring seasons over central Europe,1195

Acta Geophys., 60, 1338–1358, doi:10.2478/s11600-012-0017-9, 2012.

Preißler, J., Wagner, F., Pereira, S. N., and Guerrero-Rascado, J. L.:Multi-instrumental observation of an exceptionally strong Saha-ran dust outbreak over Portugal, J. Geophys. Res., 116, D24 204,1200

2011.Pérez, C., Nickovic, S., Baldasano, J. M., Sicard, M., Rocaden-

bosch, F., and Cachorro, V. E.: A long Saharan dust event over thewestern Mediterranean: Lidar, Sun photometer observations, andregional dust modeling, J. Geophys. Res. Atmos., 111, 2006a.1205

Pérez, C., Nickovic, S., Pejanovic, G., Baldasano, J. M., andÖzsoy, E.: Interactive dust-radiation modeling: A step to im-prove weather forecasts, J. Geophys. Res., 111, D16 206,doi:10.1029/2005JD006717, 2006b.

Pérez, C., Haustein, K., Janjic, Z., Jorba, O., Huneeus, N., Bal-1210

dasano, J. M., Black, T., Basart, S., Nickovic, S., Miller, R. L.,Perlwitz, J. P., Schulz, M., and Thomson, M.: Atmosphericdust modeling from meso to global scales with the onlineNMMB/BSC-Dust model – Part 1: Model description, annualsimulations and evaluation, Atmos. Chem. Phys., 11, 13 001–1215

13 027, 2011.Rosenfeld, D., Rudich, Y., and Lahav, R.: Desert dust suppressing

precipitation: A possible desertification feedback loop, PNAS,98, 5975–5980, 2001.

Seigneur, C., Pun, B., Pai, P., Louis, J.-F., Solomon, P., Emery,1220

C., Morris, R., Zahniser, M., Worsnop, D., and Koutrakis, P.:


Guidance for the performance evaluation of three-dimensionalair quality modeling systems for particulate matter and visibility,J. air waste manage., 50, 588–599, 2000.

Sekiyama, T. T., Tanaka, T. Y., Shimizu, A., and Miyoshi, T.: Data1225

assimilation of CALIPSO aerosol observations, Atmos. Chem.Phys., 10, 39–49, 2010.

Shao, Y., Raupach, M. R., and Findlater, P. A.: Effect of saltationbombardment on the entrainment of dust by wind, J. Geophys.Res., 98, 12 719–12 726, doi:10.1029/93JD00396, 1993.1230

Simmons, A. J. and Burridge, D. M.: An energy and angular-momentum conserving vertical finite-difference scheme and hy-brid vertical coordinates, Mon. weather rev., 109, 758–766, 1981.

Sokolik, I. N. and Toon, O. B.: Direct radiative forcing by anthro-pogenic airborne mineral aerosols, Nature, 381, 681–683, 1996.1235

Solomos, S., Kallos, G., Kushta, J., Astitha, M., Tremback, C.,Nenes, A., and Levin, Z.: An integrated modeling study on theeffects of mineral dust and sea salt particles on clouds and pre-cipitation, Atmos. Chem. Phys., 11, 873–892, 2011.

Sugimoto, N., Shimizu, A., Matsui, I., Uno, I., Arao, K., Dong, X.,1240

Zhao, S., Zhou, J., and Lee, C.-H.: Study of Asian dust phenom-ena in 2001–2003 using a network of continuously operated po-larization lidars, Water, Air, & Soil Pollution: Focus, 5, 145–157,2005.

Takamura, T. and Nakajima, T.: Overview of SKYNET and its ac-1245

tivities, Optica pura y aplicada, 37, 3303–3308, 2004.Taylor, K. E.: Summarizing multiple aspects of model performance

in a single diagram, Journal of Geophysical Research: Atmo-spheres, 106, 7183–7192, doi:10.1029/2000JD900719, 2001.

Tegen, I. and Fung, I.: Modeling of mineral dust in the atmosphere:1250

Sources, transport, and optical thickness, Journal of GeophysicalResearch, 99, 22 897, 1994.

The EARLINET publishing group 2000-2010, Adam, M., Alados-Arboledas, L., Althausen, D., Amiridis, V., Amodeo, A., Ans-mann, A., Apituley, A., Arshinov, Y., Balis, D., Belegante, L.,1255

Bobrovnikov, S., Boselli, A., Bravo-Aranda, J. A., Bösenberg,J., Carstea, E., Chaikovsky, A., Comerón, A., D’Amico, G.,Daou, D., Dreischuh, T., Engelmann, R., Finger, F., Freuden-thaler, V., Garcia-Vizcaino, D., García, A. J. F., Geiß, A., Gi-annakaki, E., Giehl, H., Giunta, A., de Graaf, M., Granados-1260

Muñoz, M. J., Grein, M., Grigorov, I., Groß, S., Gruening, C.,Guerrero-Rascado, J. L., Haeffelin, M., Hayek, T., Iarlori, M.,Kanitz, T., Kokkalis, P., Linné, H., Madonna, F., Mamouriat,R.-E., Matthias, V., Mattis, I., Menéndez, F. M., Mitev, V.,Mona, L., Morille, Y., Muñoz, C., Müller, A., Müller, D., Navas-1265

Guzmán, F., Nemuc, A., Nicolae, D., Pandolfi, M., Papayan-nis, A., Pappalardo, G., Pelon, J., Perrone, M. R., Pietruczuk,A., Pisani, G., Potma, C., Preißler, J., Pujadas, M., Putaud, J.,Radu, C., Ravetta, F., Reigert, A., Rizi, V., Rocadenbosch, F.,Rodríguez, A., Sauvage, L., Schmidt, J., Schnell, F., Schwarz,1270

A., Seifert, P., Serikov, I., Sicard, M., Silva, A. M., Simeonov,V., Siomos, N., Sirch, T., Spinelli, N., Stoyanov, D., Talianu,C., Tesche, M., De Tomasi, F., Trickl, T., Vaughan, G., Volten,H., Wagner, F., Wandinger, U., Wang, X., Wiegner, M., andWilson, K. M. a.: EARLINET all observations (2000-2010),1275

doi:10.1594/WDCC/EN_all_measurements_2000-2010, 2014.Todd, M. C., Washington, R., Martins, J. V., Dubovik, O., Lizcano,

G., M’bainayel, S., and Engelstaedter, S.: Mineral dust emis-sion from the Bodélé Depression, northern Chad, during BoDEx2005, J. Geophys. Res., 112, D06 207, 2007.1280

Figure 1. Map of the ACTRIS/EARLINET remote sensing stationsproviding data for testing the proposed methodology.

Tsekeri, A., Amiridis, V., Kokkalis, P., Basart, S., Chaikovsky, A.,Dubovik, O., Papayannis, R., Baldasano, J. M., Gross, B., andothers: Application of a synergetic lidar and sunphotometer al-gorithm for the characterization of a dust event over Athens,Greece., British Journal of Environment and Climate Change, 3,1285

531–546, 2013.Wagner, J., Ansmann, A., Wandinger, U., Seifert, P., Schwarz, A.,

Tesche, M., Chaikovsky, A., and Dubovik, O.: Evaluation of theLidar/Radiometer Inversion Code (LIRIC) to determine micro-physical properties of volcanic and desert dust, Atmos. Meas.1290

Tech., 6, 1707–1724, 2013.White, B. R.: Soil transport by winds on Mars, J. Geophys. Res.,

84, 4643–4651, doi:10.1029/JB084iB09p04643, 1979.Wiegner, M., Madonna, F., Binietoglou, I., Forkel, R., Gasteiger, J.,

Geiß, A., Pappalardo, G., Schäfer, K., and Thomas, W.: What1295

is the benefit of ceilometers for aerosol remote sensing? Ananswer from EARLINET, Atmos. Meas. Tech., 7, 1979–1997,doi:10.5194/amt-7-1979-2014, 2014.

Yumimoto, K., Uno, I., Sugimoto, N., Shimizu, A., Hara, Y., andTakemura, T.: Size-resolved adjoint inversion of Asian dust, Geo-1300

phys. Res. Lett., 39, 2012.Zhang, L., Gong, S., Padro, J., and Barrie, L.: A size-segregated

particle dry deposition scheme for an atmospheric aerosolmodule, Atmos. Environ., 35, 549–560, doi:10.1016/S1352-2310(00)00326-5, 2001.1305


� � � � ��

��

�

�

�

�

�

��

��

��

��

��

��

��

Figure 2. Distribution of available measurements per month.

� ��

��

�

�

�

�

�

��

��

Figure 3. Number of used remote sensing profiles per altitude.


� � � � � � � � �

��

�

�

�

�

�

�

�

�

�

��

��

��

��

� � � � � � � � �

��

�

�

�

�

�

�

�

�

�

��

��

��

��

� � � � � � � � �

��

�

�

�

�

�

�

�

�

�

��

��

��

��

� � � � � � � � �

��

�

�

�

�

�

�

�

�

�

��

��

��

��

Figure 4. Comparison of dust center of mass for the four models against the ones retrieved from LIRIC.

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

Figure 5. Comparison of integrated dust concentration for the four models against the ones retrieved from LIRIC. The value is calculatedonly for the altitude ranges that both models and measurements provide valid values.


� ��

��

��

�

��

��

��

��

��

��

��

��

��

��

� ��

��

��

�

��

��

��

��

��

��

��

��

��

��

� ��

��

��

�

��

��

��

��

��

��

��

��

��

��

� ��

��

��

�

��

��

��

��

��

��

��

��

��

��

Figure 6. Comparison of profile peak value for the four models against the ones retrieved from LIRIC.

��

��

��

��

��

��

��

�

��

��

��

��

��

��

��

��

�

��

��

��

��

��

��

��

��

�

��

��

��

��

��

��

��

��

�

��

Figure 7. Scatter plot of vertical correlation and fractional gross error. Black dots represent the ideal performance (0, 1). Each point on theplot corresponds to a pair consisting of one LIRIC and one model profile.The bars on the axis indicate the univariate distribution of the datafor each variable.


� ��

��

�

�

�

�

�

��

��

��

��

��

��

��

Figure 8. Average profile comparison as simulated by four modeland measured by LIRIC.

� ��

��

�

�

�

�

�

��

��

��

��

��

��

��

��

� ��

��

�

�

�

�

�

��

��

��

��

��

��

Figure 9. Comparison of average profiles simulated by all fourmodel for low and high concentration cases, separated at 0.3g/m2

��

��

�

�

�

�

�

��

��

��

��

��

��

Figure 10. Profiles of mean bias error for all four models. Grayshading indicates altitude ranges with less than 20 profiles available.

� ��

��

�

�

�

�

�

��

��

��

��

��

��

Figure 11. Profiles of root mean square error for all four models.Gray shading indicates altitude ranges with less than 20 profilesavailable.


��

�

�

�

�

�

�

��

��

��

��

��

��

Figure 12. Profiles of correlation coefficient r for all four models.Gray shading indicates altitude ranges with less than 20 profilesavailable.

��

��

�

�

�

�

�

��

��

��

��

��

��

Figure 13. Profiles of fractional gross error r for all four models.Gray shading indicates altitude ranges with less than 20 profilesavailable.


��

��

��

��

��

��

��

��

��

� � ��

�

�

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

� ��

�

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

� ��

�

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

��

� � ��

�

�

��

��

��

��

��

��

��

��

��

��

��

��

��

�

��

��

��

��

Figure 14. Taylor’s diagrams for four different altitude ranges, from 1 to 5km. The black dots represent the observational data. The distancefrom any point from the origin indicates the standard deviation of the dataset. The angular distance of a point from the horizontal axisindicates the correlation of model and the measured data. The distance of two points in these plots is proportional to their centered root meansquare (RMS) difference. L: LIRIC observations, M1: BSC/DREAM8bV2, M2: BSC/NMMB-DUST, M3: DREAMBOL, M4: DREAM8-NMME-MACC.


� ��

��

�

�

�

�

�

��

��

��

��

��

��

��

��

� ��

��

�

�

�

�

�

��

��

��

��

��

��

Figure 15. Comparison of mean concentration profiles for the west and east station clusters.

��

�

�

�

�

�

�

��

��

��

��

��

��

��

��

�

�

�

�

�

�

��

��

��

��

��

��

Figure 16. Comparison of correlation coefficient profiles for the west and east station clusters. Gray shading indicates altitude ranges withless than 20 profiles available.


Table 1. Comparison of lidar dust products for dust model evaluation.

Product Geometrical properties Quantitative Mixed dust cases Direct comparison

Range corrected signals Yes No No NoOptical products Yes Yes No NoDust optical products Yes Yes Yes NoDust concentration Yes Yes Yes Yes

Table 2. Summary of the main parameters of the dust transport models used in this study (adapted from Benedetti et al. (2014)).

BSC/DREAM8bV2 NMMB/BSC-DUST DREAMABOL DREAM8-NMME-MACC

Institution BSC-CNS BSC-CNS CNR-ISAC SEEVCCCMeteorological driver Eta/NCEP NMMB/NCEP BOLAM NMME/NCEPInitial and boundary conditions NCEP/GFS NCEP/GFS NCEP/GFS ECMWFResolution 0.33◦× 0.33◦ 0.33◦× 0.33◦ 0.4◦× 0.4◦ 0.25◦× 0.25◦

Emission scheme Uplifting–Shao et al. (1993)–Janjic (1994)–Nickovic et al. (2001)

Saltation and sand-blasting–White (1979)–Marticorena andBergametti (1995)–Janjic (1994)–Nickovic et al. (2001)

Uplifting–Shao et al. (1993)–Nickovic et al. (2001)

Uplifting–Shao et al. (1993)–Janjic (1994)–Nickovic et al. (2001)

Deposition scheme Dry deposition–Zhang et al. (2001)Below-cloud scaveng-ing–Nickovic et al. (2001)

Dry deposition–Zhang et al. (2001)Wet deposition–Ferrier et al. (2002)–Betts (1986)–Janjic (1994)

Dry deposition–Zhang et al. (2001)In and below-cloudscavenging–Maurizi and MontiConvective clouds,precipitation andre-evaporation

Dry deposition–Zhang et al. (2001)Below-cloud scaveng-ing–Nickovic et al. (2001)

Vertical resolution 24 Eta-layers 40 σ-hybrid layers 50 σ-hybrid layers 24 σ-hybrid layersRadiation interaction Yes No No NoData assimilation No No No Yes


Table 3. Definition, value range, and ideal score of metrics for the systematic evaluation of modeled dust concentration profiles. c denotesthe concentration at altitude z. Mi and Oi represent modeled and observed profiles respectively for the ith measurement pair. Altitudedependence is omitted for brevity.

Metric Definition Range Perfect score

Center of mass

� zmax

zminz · c · dz

� zmax

zminc · dz - -

Mean Bias

�1

N

N�

i=1

(Mi −Oi)2

� 12

∞ to ∞ 0

Correlation coefficient�N

i=1

�Mi −M

��Oi −O

��N

i=1

�Mi −M

�2�Ni=1

�Oi −O

�2� 12

-1 to 1 1

Root Mean Square Error

�1

N

N�

i=1

(Mi −Oi)2

� 12

0 to ∞ 0

Mean Fractional Bias2

N

N�

i=1

�Mi −Oi

Mi +Oi

�-2 to 2 0

Fractional Gross Error2

N

N�

i=1

��Mi −Oi

Mi +Oi

�� 0 to 2 0

Table 4. The following 10 stations provided dust profiles retrieved by the combined lidar and AERONET photometer. The provided referencesgive further information for each station and the measurement instruments.

Station Location (oN,oE) Altitude (m) Lidar channels No. Profiles Reference

Athens 37.97, 23.77 200 3β 3 Kokkalis et al. (2012)Barcelona 41.39, 2.17 115 3β 9 Kumar et al. (2011)Belsk 51.84, 20.79 180 3β 1 Pietruczuk and Chaikovsky (2012)Bucharest 44.35, 26.03 93 3β+1δ 5 Nemuc et al. (2013)Evora 38.57, -7.91 293 3β+1δ∗ 21 Preißler et al. (2011)Granada 37.16, -3.61 680 3β+1δ 11 Guerrero-Rascado et al. (2009)Lecce 40.30, 18.10 30 3β+1δ 4 Perrone et al. (2014)Leipzig 37.97, 23.77 90 3β+1δ 3Potenza 51.35, 12.44 760 3β+1δ 8 Madonna et al. (2011)Thessaloniki 40.63, 22.95 60 3β 4 Papayannis et al. (2012)

Table 5. Correlation coefficient (r) and fractional bias (FB) for single value metrics of the compared profiles.

Center of Mass Total concentration Peak valuer FB r FB r FB

BSC/DREAM8bV2 0.37 0.00 0.54 -0.90 0.38 -0.90NMMB/BSC-DUST 0.57 0.16 0.64 -0.74 0.34 -0.78DREAMABOL 0.53 0.00 0.24 -1.04 0.19 -0.90DREAM8-NMME-MACC 0.62 0.12 0.58 -0.21 0.48 -0.31

Documents

Application of the Lidar/photometer algorithm: dust model ...Manuscript prepared for Atmos. Meas. Tech. with version 2014/07/29 7.12 Copernicus papers of the LATEX class coperni- cus.cls