Embed Size (px)
Citation preview
ARTICLE IN PRESS
1352-2310/$ - se
doi:10.1016/j.at
�Correspondfax: +331 64 1
E-mail addr
Atmospheric Environment 41 (2007) 5300–5315
www.elsevier.com/locate/atmosenv
Validation of the POLYPHEMUS platform on the ETEX,Chernobyl and Algeciras cases
Denis Queloa, Monika Krystaa, Marc Bocqueta, Olivier Isnardb,Yannick Minierb, Bruno Sportissea,�
aCLIME, Joint INRIA/ENPC Project, CEREA, Joint Laboratory Ecole des Ponts/EDF R&D, FrancebInstitute of Radiation Protection and Nuclear Safety, BP 17, 92262, Fontenay-aux-Roses, France
Received 6 July 2006; received in revised form 9 January 2007; accepted 22 February 2007
Abstract
The objective of this article is to investigate the validity of a modeling system developed for forecasting atmospheric
dispersion, the POLYPHEMUS platform, with a special focus on radionuclides. The platform is briefly described and model-
to-data comparisons are reported for three cases: the ETEX campaign, the Chernobyl accident and the Algeciras release.
The results are similar to those usually given in the literature by state-of-the-art models. Some preliminary sensitivity
analysis indicates the main sources for uncertainties.
r 2007 Elsevier Ltd. All rights reserved.
Keywords: Radionuclides; Chernobyl; Algeciras; ETEX; Air quality modeling; POLYPHEMUS; POLAIR3D; Long-range transport
1. Introduction
A specific field of air quality modeling is relatedto risk assessment of accidental industrial releases.This concerns point emissions (local in space andtime) of trace species that have zero or lowatmospheric concentrations. In the late 1980s andearly 1990s, considerable efforts have been devotedto the development of operational models, espe-cially after the Chernobyl accident. There are nowmany available models ranging from Gaussian-likemodels for short-range dispersion (and operationalemergency centers, for instance (Pasler-Sauer, 2003;Thykier-Nielsen et al., 1999)) to Lagrangian models
e front matter r 2007 Elsevier Ltd. All rights reserved
mosenv.2007.02.035
ing author. Tel.: +33 1 64 15 21 41;
5 21 70.
ess: [email protected] (B. Sportisse).
and three-dimensional Eulerian models for long-range transport (see below).
Many models, based on state-of-the-art Chemis-try-Transport Models (the so-called CTMs) havebeen developed, used or updated in this context(dispersion of radionuclides at regional scale).Model-to-data comparisons have been performedwith a small set of observational data (typically theChernobyl case and the ETEX campaign). Somemodel intercomparisons have also been performed(the ATMES exercise for instance, Klug et al.(1992)). Among many other models, we can forinstance cite the NAME (Maryon et al., 1992),EURAD (Haas et al., 1990) or DREAM (Brandtet al., 2002).
Modern numerical systems are supposed to havemany features, forecasting being the first of all.Many other high-level applications are also
.
ARTICLE IN PRESSD. Quelo et al. / Atmospheric Environment 41 (2007) 5300–5315 5301
required, especially for risk assessment: getting anestimation of uncertainties through ensemble mod-eling or Monte Carlo simulations, inverse modelingof unknown emissions, data assimilation of ob-servational data, etc. As such, the numerical modelis only a ‘‘small’’ component of such systems.
The POLYPHEMUS platform has been developed inthis context. The key motivation is a clear partition-ing between different tasks: parameterizations andpreprocessing of meteorological fields, model dri-vers (for forecast, for data assimilation, for en-semble simulations), models (especially the CTMPOLAIR3D (Boutahar et al., 2004)) and postproces-sing tools (for instance for ensemble statistics). ThePOLYPHEMUS system has already been used for manyapplications with a focus on air quality: sensitivityanalysis of ozone with respect to emissions (Malletand Sportisse, 2005), evaluation of uncertainties(Mallet and Sportisse, 2006b), inverse modeling ofNOx emissions at regional scale (Quelo et al.,2005a, b), ensemble forecast for ozone (Mallet andSportisse, 2006a), modeling of mercury and heavymetals at continental scale (Roustan and Bocquet,2006a, b), etc.
The POLYPHEMUS platform will be operational atthe Crisis Center of the French Institute ofRadiological Protection and Nuclear Safety in thebeginning of 2007. It will be a component of amodeling system KRX devoted to long-range (theLDX system) and short-range (the PX system)atmospheric transport and radiological dose evalua-tions. In this configuration, POLYPHEMUS will becoupled to a meteorological server, which willprovide operational forecasts from Meteo Franceand ECMWF and detailed forecast with the MM5mesoscale model.
The objective of this paper is to report validationsof the POLYPHEMUS platform applied to the disper-sion of radionuclides. Preliminary applications havebeen performed by former versions of the POLAIR3Dmodel (for instance in Issartel (2003); Bocquet(2005b) towards the ETEX data). As in Baklanovand Sorensen (2001), the case studies are the ETEXcampaign, the Chernobyl accident and the Algecirasrelease.
This paper is organized as follows. Section 2briefly describes the POLYPHEMUS platform with afocus on dispersion of radionuclides. Sections 3, 4and 5 detail the applications to the ETEX cam-paign, the Chernobyl accident and the Algecirasrelease. Model-to-data comparisons are performedon the basis of available observational data. Some
preliminary sensitivity analyses are also performedin order to underline the robustness and thelimitations of the system. The article ends with aconclusion. Some future planned developments arealso summarized with a focus on aerosol modeling,data assimilation and ensemble techniques.
2. Description of the POLYPHEMUS platform
2.1. Structure
The POLYPHEMUS platform is made of fourcomponents:
�
Physical parameterizations and preprocessing ofinput fields (meteorological fields, boundaryconditions and emissions) are performed withthe ATMO-DATA library. These fields are given forthe grid of the model (in the case of this article:the CTM POLAIR3D). � Drivers for high-level uses of models (driver forMonte Carlo simulations, driver for ensembleforecast (similar to Galmarini et al. (2004a, b)),driver for sequential data assimilation, driver forvariational data assimilation (Robertson andPerrson, 1993; Krysta et al., 2004; Quelo et al.,2005a, b)) have been defined in order to handlehigh-level uses of models. These drivers do notdepend on the chosen model. Part of this level isstill a work in progress.
� Models (especially the CTM POLAIR3D (Bouta-har et al., 2004)) constitute the third level of thesystem.
� The last level is composed of postprocessingtools, especially with the Python module ATMO-
PY. This library (which does not depend on theprevious components) performs model-to-datacomparisons.
2.2. The POLAIR3D CTM
The POLAIR3D CTM is basically a numericalsolver of the reactive dispersion equation for theconcentration of the trace species ci (labeled with i)
qci
qtþ divðVrciÞ ¼ div rKr
ci
r
� �� �� Ls
i ci � Ldi ci þ Si ð1Þ
with Lsi the wet scavenging coefficient, Ld
i theradioactive decay and Si the point source for species
ARTICLE IN PRESSD. Quelo et al. / Atmospheric Environment 41 (2007) 5300–53155302
i. K is the eddy diffusivity matrix, supposed to bediagonal. The vertical component is given by Kz (thedefault parameterization is the Louis parameteriza-tion (Louis, 1979)). The horizontal component has aconstant value KH (in this study, KH ¼ 0).
Some boundary conditions have to be specified,especially at ground
�Kzrci � n ¼ Ei � vdepi ci, (2)
where n is the upward oriented unitary normalvector and Ei is the surface emissions for species i (0in this case).
POLAIR3D solves the dispersion equation with anoperator splitting method with many availableoptions. In this specific case the source term islinear so there is no impact of the splitting strategy.
The advection step is solved with a third-orderDirect Space Time scheme with a Koren–Swebylimiter function. The three directions are solvedsimultaneously. Because of the sharp gradientsgenerated by the point-wise source, it is importantthat the advection scheme relies on a flux limiter. Thediffusion step is solved with an implicit model(a second-order Rosenbrock method) with a three-points stencil for spatial discretization. Directionalsplitting is used. The scavenging step and the radio-active decay step are solved analytically. The groundboundary conditions are added to the diffusion step.
We refer to Boutahar et al. (2004) for a deeperdescription of POLAIR3D. A general overview ofalgorithms is given in Verwer et al. (1998) and acomprehensive investigation of numerics in PO-
LAIR3D may be found in Mallet et al. (in revision).
2.3. Some specific parameterizations
In these first applications, the only radionuclidestaken into account are cesium and iodine. Twoisotopes of cesium are modeled: 134Cs and 137Cs.Iodine is known to be in many possible forms(elemental gas-phase iodine, organic methyliodineor particle-bound iodine). We only consider onelumped form for 131I. In order to be able to evaluateradiological consequences (doses), a comprehensiveset of radionuclides will be added in the future.
There are two modes of the model configurationfor the dispersion of radionuclides:
�
in a first mode, the trace species are supposed toact as gases; � in a second mode, some trace species act asmonodisperse aerosols. As such, they have a
gravitational settling velocity and specific scaven-ging properties as functions of the particle size.
POLAIR3D may host two aerosol models: MAM, aModal Aerosol Model and SIREAM, a SizeResolved Aerosol Model. These models solve theGeneral Dynamics Equation for aerosols, by takinginto account Brownian coagulation, condensation/evaporation and nucleation. Due to the lack of datafor radionuclides, these models are not used for thisapplication even if the attachment to atmosphericaerosols is known to be a key process for long-rangetransport (Pollanen et al., 1997).
Resistance models for dry deposition and micro-physical parameterizations for wet scavenging areavailable in the model. For a preliminary evaluationof the system, we have chosen to use a simpleapproach:
�
a constant dry deposition velocity vdep ¼0.2 cm s�1 for cesium, vdep ¼ 0.5 cm s�1 foriodine;
� a scavenging coefficient parameterized as Ls ¼ApB0 with A ¼ 8� 10�5, B ¼ 0.8 and p0 the rain
intensity (in mmh�1).
We refer to Brandt et al. (2002); Baklanov andSorensen (2001); Baklanov (1999) and Sportisse(2007) for comprehensive studies of these parame-terizations.
The radioactive decay is only taken into accountfor the three trace species. The following values (ins�1) have been used:
Ld131I ¼ 9:97� 10�7; Ld
131Cs ¼ 1:08� 10�8,
Ld137Cs ¼ 7:32� 10�10 ð3Þ
which correspond to lifetimes of 8 days, 2 years and30 years for 131I, 134Cs and 137Cs, respectively.
In a future version, a comprehensive mechanismfor radioactive filiation will be added thanks tothe Bateman’s formula, which gives analyticalsolutions.
2.4. Statistical indicators
Accidental releases of pollutants are describedstatistically quite differently from diffuse pollutants(passive or reactive). This is mainly due to theobvious statement that the pollutant forms a clear-cut cloud in the short term. As a consequence theconcentration profile at a given station is made up
ARTICLE IN PRESSD. Quelo et al. / Atmospheric Environment 41 (2007) 5300–5315 5303
of smoothed peaks giving away the cloud passings.Therefore statistical indicators measuring the skillsof a model simulating releases should be adequate.Specifically classical indicators such as correlationcoefficients, bias, figure of merit, normalized meansquare error, should not only apply to concentra-tions, maximum concentrations, or dosage (time-integrated concentrations), but also to the followingrelevant observables, as introduced in Brandt et al.(1998):
�
The arrival time of the cloud which is the firsttime the concentration in tracer exceeds thebackground value threshold. � The arrival time of maximum concentration whichmeasures time when the maximum concentrationat the station occurs.
� The duration coefficient which is the time spent bythe concentration at a given station above athreshold.
Qualitative global scatter diagrams can be usedon those observables, as well as the classicalstatistical indicators mentioned above.
To evaluate the agreement between measured andpredicted tracer concentration values in time, astatistical methodology has been designed for theETEX intercomparison exercises. A comprehensivedescription of the indicators, which were applied tothe simulations can be found in Appendix.
3. Application to the European Tracer Experiment
(ETEX) campaign
The ETEX is one of the best-instrumented disper-sion experiment at continental scale to date. More-over, it is well documented ((JRC, 1998), and onlinedocumentation: http://rem.jrc.cec.eu.int/etex/).
As for phase I of ETEX, 340 kg of perfluoro-methylcyclohexane (PMCH) were released uni-formly from 23/10/1994 1600UTC to 24/10/19940350UTC, at Monterfil (Brittany, France, located481030N, 21000W). This compound was chosenbecause it is safe, inert, and is hardly removed bywet scavenging or dry deposition.
Total of 168 WMO stations over 17 countries,owned by the local meteorological services, partici-pated in the observation campaign, yielding 969measurements of PMCH concentrations abovePMCH background level and 2136 others (qualitycontrol checked) concentrations within the back-ground PMCH noise level. The results of the
experiment were used to calibrate various atmo-spheric dispersion models. In particular, theATMES II intercomparison exercise (43 partici-pants) focused on ETEX-I. Many modeling andcomparison to data results were also collected in thespecial issue of Atmospheric Environment 32(issue24), 4089–4375, 1998. It was the central topic of thelong-range transport, model verification and emer-gency response which took place in Vienna in 1997(Nodop, 1997).
Additionally these measurements were used totest source inversion methods on a real and well-instrumented dispersion event. Among the contri-butions on the subject (Robertson and Langner,1997, 1998; Issartel and Baverel, 2003; Issartel,2003; Bocquet, 2005a, b), several were using PO-
LAIR3D, the CTM of the POLYPHEMUS system.
3.1. Set-up
For the simulation presented here, the domainwas chosen large enough so that outgoing PMCH isnot significantly tampering the simulation for atleast several days. The center of the south-westcorner cell is at �201W, 36.051N. It was chosenso that the ETEX release point be at the center ofa cell.
The finest meteorological fields were providedby the ECMWF at the resolution 0.51� 0.51. Theseare operational fields. For the purpose of thesimulation ahead, we use ERA-40 fields, with alower resolution of 1.1251� 1.1251, but a greaterconfidence (these are re-analyzed fields). The time-step is 3 h.
The resolution of the simulation may differ fromthe meteorological fields resolution (the meteorolo-gical fields are then interpolated). A finer nestedLagrangian model operating in the vicinity of thesource may make a difference. For the simulationahead, the resolution is chosen to be 0.51� 0.51 witha time step Dt ¼ 600 s. The number of cells isNx ¼ 137 and Ny ¼ 73.
For this simulation, 12 vertical levels are takeninto account, up to 6090m.
3.2. Results and discussion
We have evaluated our simulation against theobservations at stations with a minimum of 11measurements (enough to infer statistics on arrivaltime, maximum concentrations, etc.). There are 139such stations. On Fig. 1, we have plotted the scatter
ARTICLE IN PRESS
Table 1
Statistical indicators for the dosage, arrival time, duration,
maximum concentration and arrival time of maximum concen-
trations of the ETEX-I release
Observable vs. indicator Correlation FM NMSE FB
Dosage 0.80 0.32 6.27 0.93
Arrival time 0.94 0.90 0.13 �0.023
Duration 0.74 0.66 0.54 0.11
Maximum 0.83 0.37 3.84 0.75
Peak time 0.58 0.77 0.92 0.15
D. Quelo et al. / Atmospheric Environment 41 (2007) 5300–53155304
plots for dosage, arrival time, duration (total timeabove 0.05 ngm�3), maximum concentration andarrival time of maximum concentrations.
Firstly, the performance of the model on thestatistics of observables built on the concentrationswas investigated. The statistical indicators whichhave been computed for these observables are givenin Table 1.
When considering all concentration measure-ments of the same 139 stations, the correlation is0.60, the FM is 0.26, the NMSE is 5.00, the FB is0.89. When considering all measurements, thecorrelation is 0.58, the FM is 0.28, the NMSEis 3.98, the FB is 0.81. These statistics comparewell with state-of-the-art models (Brandt, 1998).However the model is incline to overestimate the
Fig. 1. Scatter plots of the dosage (top left) arrival time (top right), m
computed on the set of 139 stations of the ETEX-I release.
station dosages of PMCH (the fractional bias being0.93). Following the ATMES II statistical metho-dology, other indicators for concentrations havebeen computed. Global analysis indicates good
aximum concentration (bottom left) and duration (bottom right)
ARTICLE IN PRESS
20 30 40 50 60 70
Hours after the release start
0.0
0.2
0.4
0.6
0.8
1.0
FM
S
Fig. 2. Time evolution of the FMS for the ETEX-I release.
Table 2
Statistical indicators for the concentrations for the ETEX-1
exercise
Station FMT NMSE MBE Correlation
First arc
B05 0.13 32.2 �0.33 0.90
D44 0.35 7.06 �0.19 0.96
NL01 0.39 5.23 �0.19 0.73
NL05 0.19 7.09 0.05 0.50
Second arc
CR03 0.53 2.41 �0.04 0.96
D05 0.69 0.51 �0.04 0.97
D42 0.41 2.71 0.16 0.76
DK02 0.76 0.50 �0.04 0.98
DK05 0.41 2.40 0.10 0.86
H02 0.64 0.84 �0.01 0.93
PL03 0.53 1.70 �0.06 0.88
D. Quelo et al. / Atmospheric Environment 41 (2007) 5300–5315 5305
agreement with the observations for FA2 (0.73%)and FA5 (0.80%). Fig. 2 displays the time evolutionof the FMS computed for all available measure-ments. The results of this indicator are satisfactory.For example, at 36 h, the FMS is 55% whereas inATMES-II, a quarter of the models had a FMS455% and the maximum FMS was 71%. Timeanalysis is illustrated in Table 2, which summarizedsome statistical indicators for the stations charac-terized by arrival times between 15 and 18 h (firstarc) or equal to or beyond 30 h (second arc) after therelease start.
4. Application to the Chernobyl accident
An explosive accident took place at the Cherno-byl nuclear power plant unit IV in Ukraine(511170N, 301150E) on 25 April 1986 at 2123UTC.This accident led to a widespread dispersion ofradionuclides in the atmosphere at the continentalscale. Most of the released material was inparticulate form except for noble gases and themajority of iodine. Radioactivity was measured inmany European countries after the accident.
The measured data are provided by the REM-database at the Environment Institute (Joint Re-search Center, Ispra, Italy). They consist of datafrom 88 stations. This represents up to 1278observational data for cesium and 1333 observa-tional data for iodine (whatever the measurementduration is). The map of stations is given in Fig. 3(bottom right).
4.1. Set-up
Contrary to the ETEX campaign, the estimate ofthe emissions are still highly uncertain: this concernsthe total released activity, the time distribution(most of the release occurred during the period 25April–5 May) and the vertical distribution. Notethat the estimation of the source term has beenseveral times corrected by more than a factor of twoand that Brandt et al. (2002) reports an uncertaintyof at least 50%.
The key uncertainty is related to the verticaldistribution of the emissions: due to the hightemperature of the core, the material was assumedto reach heights up to 2000m, probably more. Theeffective release height changed considerably duringthe release: the initial explosion lifted material tohigh altitude while release heights were probablymuch lower in the next 2 weeks.
In our runs we have chosen to follow therecommendations of UNSCEAR (2000) for thetime distribution and the total released activity, andof Brandt et al. (2002) for the vertical distribution.The time evolution and the vertical distribution wehave used are given in Table 3.
Even if the radionuclide composition is known tohave been very complex, we have chosen to describeonly iodine (131I) and cesium (with the two isotopes134Cs and 137Cs).
Another key uncertainty is the partitioningamong particulate phase and gaseous form. Werefer to Poollanen et al. (1997) for a comprehensive
ARTICLE IN PRESS
Table 3
Time and vertical distribution for Chernobyl emission
26/4 27/4 28/4 29/4 30/4 1/5 2/5 3/5 4/5 5/5
t 0.40 0.116 0.085 0.058 0.039 0.035 0.058 0.061 0.074 0.074
z1 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0
z2 0.0 0.0 0.2 0.2 0.2 0.2 0.2 0.2 0.2 0.2
z3 0.0 0.1 0.5 0.5 0.5 0.5 0.5 0.5 0.5 0.5
z4 0.0 0.5 0.3 0.3 0.3 0.3 0.3 0.3 0.3 0.3
z5 0.4 0.4 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0
z6 0.3 0.0 0. 0.0 0.0 0.0 0.0 0.0 0.0 0.0
z7 0.2 0.0 0. 0.0 0.0 0.0 0.0 0.0 0.0 0.0
z8 0.1 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0
The levels zi are: 32, 150 360, 640, 990, 1400, 1800 and 2300m.
Fig. 3. Chernobyl accident: maps of ground concentrations of 137Cs (in Bqm�3) at 1300UTC on 26 April (top left), 29 April (top right)
and 2 May (bottom left). The map of stations is given bottom right.
D. Quelo et al. / Atmospheric Environment 41 (2007) 5300–53155306
study of the radioactive particles from the Cherno-byl accident. The impact of large particles (aero-dynamic diameters larger than 20 mm) isinvestigated. Another interesting feature pointedout by this article is the trade-off between effectiverelease height higher than usually reported andvertical upwards motions due to convective cellswith rising currents of warm air (especially during
April 26 over Belarus and Baltic states). Notice thatwe have not included the description of convectivemotions in our dispersion model.
Weather conditions during the release are welldocumented. A high-pressure center with stronginversion at about 500m was active over the site.
The meteorological data used as inputs for thePOLYPHEMUS system are ERA-40 data from
ARTICLE IN PRESSD. Quelo et al. / Atmospheric Environment 41 (2007) 5300–5315 5307
ECMWF with a 1.1251� 1.1251 horizontal resolu-tion. The time resolution is 3 h and there are 61vertical levels (hybrid coordinates).
In some configurations of the runs, we have alsoused meteorological data computed by the MM5mesoscale model (version v5.2). The concern wasmainly to investigate the sensitivity with respect torain intensity and many microphysical schemesparameterizing cloud and rain processes in MM5have been tested. For the sake of clarity, the resultsare not reported here and we refer to a future paper.
The horizontal grid is made of 66� 31 cells(1.1251� 1.1251 horizontal resolution). The verticalgrid of POLAIR3D is made of 12 levels with thefollowing heights of the interfaces: 62, 236, 484, 796,1184, 1616, 1984, 2616, 3184, 3616, 4384 and5016m. The timestep is 600 s.
4.2. Results and discussion
Maps of the computed surface concentration of137Cs are given in Fig. 3 for 26, 29 April and 2 May(13:00UTC), respectively (1, 4 and 7 days after thestart of the release). The plume location is coherentwith the meteorological understanding and withsimilar maps reported in the literature.
Model-to-data comparisons are illustrated byFig. 4 with a scatter plot of 137Cs.
Specific results for each station are available onrequest (but not reported here due to the length ofthe table—more than two pages). Some typicalmodel-to-data comparisons are illustrated in Fig. 5.Error statistics are reported in Table 4.
Measurements [Bq.m{-3}]
10-3
10-3
10-2
10-2
10-1
10-1
100
100
101
101
102
102
Poly
phem
us [B
q.m
{-3} ]
Fig. 4. Chernobyl accident: results for the simulation of 137Cs.
Scatter plot for activity concentrations in logarithmic scale.
To date, the most comprehensive modelingstudies (up to our knowledge) devoted to theChernobyl accident are those of Brandt et al.(2002). These results are similar to those obtainedin these studies.
Notice that these results are particularly sensitiveto the source term. For instance, with the timedistribution used in Brandt et al. (2002), thecorrelation for iodine and cesium is 35% and38.5%.
5. Application to the Algeciras release
In late May and early June 1998, monitoringnetworks of several European countries detectedelevated levels of radioactivity in the atmosphere.The highest values have been recorded in thesouth of France and north of Italy. Several daysafterwards it was established that they resultedfrom a release that had taken place in a steelmill located in the south of Spain. The Aceri-nox furnace planted in between Algeciras andGibraltar melted accidentally a radiotherapysource of 137Cs. The release was qualified as incidentby the IAEA.
It is agreed today that it took place between 0100and 0300UTC on 30 May 1998. The esti-mated quantity of the released radioactivity isknown up to an order of magnitude [2.96� 1011,2.96� 1012] Bq.
About 118 mean activity concentrations in airhave been measured. For each measurement thedate of the beginning and the end of the measure-ment interval has been reported. No more preciseinformation has been given however. Commonsense makes one suppose it started at 1200UTCon the first day and the ended at 1200UTC on thelast day. The same hypothesis has been made inVogt et al. (1998). The measurement intervals coverperiods of between 1 and 14 days and in someplaces they start even before the release actuallytook place.
The localization of the monitoring stations isshown in Fig. 6. Half of the measurements comefrom the stations where one measurement only hasbeen taken. For the remaining half two or morevalues have been recorded. The monitoring stationsin several places in France and Italy, Fig. 8 and Fig.9, respectively, have recorded up to 5 dailymeasurements. Their values reflect the shape of thepassing cloud.
ARTICLE IN PRESS
avr 28 mai 01 mai 04 mai 070.0
0.5
1.0
1.5
2.0
2.5
3.0V
olu
mic
activity [
Bq
.m{-
3} ]
0.0
0.5
1.0
1.5
2.0
2.5
3.5
3.0
Vo
lum
ic a
ctivity [
Bq
.m{-
3} ]
Vo
lum
ic a
ctivity [
Bq
.m{-
3} ]
0.0
0.5
1.0
1.5
2.0
2.5
Vo
lum
ic a
ctivity [
Bq
.m{-
3} ]
0.0
0.5
1.0
1.5
2.0
2.5
Vo
lum
ic a
ctivity [
Bq
.m{-
3} ]
Vo
lum
ic a
ctivity [
Bq
.m{-
3} ]
AachenRWTH
avr 28 mai 01 mai 04 mai 07
0
2
4
6
8
10
12CeskeBudejovice
avr 28 mai 01 mai 04 mai 07
Harwell
avr 28 mai 01 mai 04 mai 07
Kozanis
avr 28 mai 01 mai 04 mai 070
2
4
6
8
10
12
14
16
18
MoravskyKrumlov
avr 28 mai 01 mai 04 mai 07
Trisaia
Fig. 5. Chernobyl accident: concentration of 137Cs activity expressed in Bqm�3 as measured (rectangles) and predicted by the model. The
solid line illustrates instantaneous profile and diamonds stand for model values corresponding to the measurements. From left to right and
from top to bottom the stations are: Aachen RWTH, Ceske Budejovice, Harwell, Kozanis, Moravsky Krumlov, Trisaia.
D. Quelo et al. / Atmospheric Environment 41 (2007) 5300–53155308
5.1. Set-up
To our knowledge, the first attempt to simulateconsequences of the incident came from theLawrence Livermore National Laboratory. On the
basis of model-to-data comparison (Vogt et al.,1998) attempted to establish the time of the releaseand the rejected activity. Buckley (1999) and, to alarger extent, Baklanov (1999) focused on theinfluence of refinement of dry and wet deposition
ARTICLE IN PRESS
Table 4
Error statistics for Chernobyl accident
Species Obs.
data
Obs.
mean
Sim.
mean
Correlation
(%)
NMSE
131I 1333 3.872 8.142 44.86 1.397137Cs 1278 1.103 1.111 45.5 1.146
Fig. 6. Monitoring stations for the Algeciras incident are drawn
with circles. Circle diameter corresponds to the value of the
largest measurement taken at a given point. Source localization in
Algeciras is marked with a triangle.
D. Quelo et al. / Atmospheric Environment 41 (2007) 5300–5315 5309
schemes. The latter report presents simulations thattook place within the RTMOD exercise (Grazianiet al., 2000).
The results of the simulation presented in thispaper have been obtained with the POLYPHEMUS
platform. The domain of computation is spanned bythe intervals [181150W, 271150E]� [261450N,591450N]. The mesh consists of 91� 66 cells at eachof 14 vertical layers. It covers the location of theaccident (51260W, 361100N) and extends over thewhole area where the measurements were collected.ECMWF operational meteorological fields withspatial resolution of 0.51� 0.51 and 6 h frequencyhave been used in this simulation. Following thestudies in Baklanov (1999) the time of the releasehas been chosen between 0130 and 0200UTC on 30May 1998 and the quantity of the released activityequals to 1.85� 1012 Bq. Due to the elevation of therelease point resulting from the presence of a stack,the source has been placed at the second level i.e.between 64 and 236m.
Dry deposition is parameterized with constantvelocity equal to vd ¼ 0.1 cm s�1. This value gives
better results than the default value 0.2 cm s�1,which justifies its use in this case. Due to lack ofinformation concerning rainfall in the operationalECMWF fields, scavenging has been ignored.Radioactive decay of 137Cs has been taken intoaccount.
5.2. Results and discussion
First, the development of the simulated plumecan be followed in Fig. 7. Several maps of Europepresented in this figure show the plume’s extentevery 12 h. The first one illustrates the cloud at1200UTC on 30 May 1998, 10 h after the end of therelease. The last one dates from 0000UTC on 5June 1998.
Simulation results show eastward plume advec-tion over the Mediterranean Sea in the first phaseafter the accident. It explains the scarcity ofmeasurements over the Spanish territory. Largediscrepancy between the measured and observedvalue in Gibraltar reflects significant model error inthe vicinity of the source. Further downwindhowever, along the Spanish coast, we note bettermodel-to-data correspondence. Moreover, ourmodel predicts correctly the arrival time at southernFrance and northern Italy on 2 June 1998, Figs. 8and 9. Model results for Nice match quite well themeasurements. Those for Cadarache and Montpel-lier constantly overpredict them. Next, the plumemoves towards Switzerland and Italy. The temporalprofile of the passing cloud, both its shape andconcentration values, are reproduced reasonablywell for the Italian station of Ispra, Fig. 8. Theshape of the modeled profile resembles to themeasured one for Capo Mele and Vercelli. It ishowever totally off for Trino. Two possible reasonsbehind it could be conjured. Firstly, it might be dueto a stagnant cloud that has been trapped some-where between the mountains. The second possibleexplanation is a mistake in the reported measure-ment dates. Trino lies only 20 km from Vercelli androughly half-way between Torino and Milan.Curiously enough, the concentrations that havebeen recorded there are shifted towards later timeswith respect to the surrounding stations.
Monitoring stations scattered over Central Eur-ope were swept by a well-spread and diluted cloudand consequently less severely contaminated thanthose in France or Italy. Not only were they at largedistances from the source but also collected usuallyno more than one measurement averaged over a
ARTICLE IN PRESS
Fig. 7. Plume of 137Cs for the Algeciras accident. Activity concentrations are expressed in Bqm�3.
D. Quelo et al. / Atmospheric Environment 41 (2007) 5300–53155310
period of several days. Therefore, quality of theresults for those sites is more difficult to evaluate.
Model–data comparisons for several monitoringstations are shown beneath. The selected stationshave already been discussed and are those thatcaptured the evolution of daily-averaged radio-activity concentrations.
In Fig. 10, a global correspondence between thesimulated air concentrations and the measured onesis illustrated for all 118 available measurements.The majority of the points are situated between twodotted lines. It implies that for most of themonitoring stations model predictions come withinan order of magnitude to the measurement. Thereare some however for which the discrepancies arehigher. At the first sight one notices that the pointsare roughly uniformly distributed on both sides ofthe diagonal. Quantitatively this fact is reflected in
the small values of bias, fractional bias and meangeometric bias, Table 5. The last quantity isparticularly well adapted to the analysis of twodatasets with large discrepancies like the ones forthe Algeciras simulation. The distribution of thepoints around the diagonal of the scatter-plot mightbe uniform but the discrepancies are large. There-fore the statistical indicators based on variance, likethe normalized mean square error, fractionalstandard deviation and geometric mean variancehave higher values. The correlation of 20% is quitepoor mostly due to the fact that no measurementselection has been done and all have been taken intoaccount in the validation procedure. It is enough toget rid of the two measurements in the close vicinityof the source so that is rises up to 46%.
The set-up of the simulation illustrated in Fig. 7consists in using operational meteorological fields of
ARTICLE IN PRESS
May 30 Jun 02 Jun 05 Jun 08 Jun 110
200
400
600
800
1000
1200Ispra
May 30 Jun 02 Jun 05 Jun 08 Jun 110
200
400
600
800
1000
1200
1400
1600Vercelli
May 30 Jun 02 Jun 05 Jun 08 Jun 110
500
1000
1500
2000
2500
3000
3500Capo_Mele
May 30 Jun 02 Jun 05 Jun 08 Jun 110
200
400
600
800
1000
1200
1400
1600
1800Torino
May 30 Jun 02 Jun 05 Jun 08 Jun 110
200
400
600
800
1000
1200
1400
1600
1800Trino
May 30 Jun 02 Jun 05 Jun 08 Jun 110
200
400
600
800
1000
1200
1400
1600
1800Milano
Fig. 8. Algeciras incident: concentration of activity expressed in mBqm�3 as measured (rectangles) and predicted by the model. The solid
line illustrates instantaneous profile and diamonds stand for model values corresponding to the measurements. From left to right and from
top to bottom the stations are: Ispra, Vercelli Capo Mele Torino Trino and Milano.
D. Quelo et al. / Atmospheric Environment 41 (2007) 5300–5315 5311
the ECMWF. Previously in this paper ERA-40fields have been used and their merits discussed.Nevertheless, for the dispersion of radioactivity
after the Algeciras release the simulation based onoperational fields give better results than the one forERA-40. It is clearly visible if one analysis the
ARTICLE IN PRESS
May 30 Jun 02 Jun 05 Jun 08 Jun 11
0
500
1000
1500
2000
2500
3000
Nice
May 30 Jun 02 Jun 05 Jun 08 Jun 11
0
500
1000
1500
2000
2500
3000
3500
4000
4500
Cadarache
May 30 Jun 02 Jun 05 Jun 08 Jun 11
0
1000
2000
3000
4000
5000
6000
7000
8000
Montpellier
Fig. 9. Algeciras incident: profiles of activity concentration, mBqm�3 in Nice, Cadarache and Montpellier in the south of France. The
measurements are represented with the rectangles, the solid line represents the simulated profile and diamonds stand for measures as
predicted by the model.
Measurements
100
100
101
101
102
102
103
103
104
104
Poly
phem
us
Cs137
Fig. 10. Algeciras incident: scatter plot for model-to-data
comparison in logarithmic scale.
D. Quelo et al. / Atmospheric Environment 41 (2007) 5300–53155312
values of the statistical indicators as shown inTable 5. All of the values associated with the ERA-40 fields indicate much worse model performance.
6. Conclusion
A preliminary validation of the POLYPHEMUS
system has been performed toward measured datafor the Chernobyl accident, the ETEX campaignand the Algeciras release. The statistical indicatorsfor model-to-data comparisons indicate a goodbehavior of the system as compared to other state-of-the-art models.
POLYPHEMUS will be the basis of the comingoperational system (LDX) used for long-rangetransport at IRSN (Institute of Radiation Protec-tion and Nuclear Safety). Complementary works
ARTICLE IN PRESS
Table 5
Algeciras incident: model-to-data comparison for the Algeciras
release
Indicator ERA-40 Operational
M (mBqm�3) 365.3 365.3
P (mBqm�3) 249.1 398.2
B (mBqm�3) �116.2 328.5
FB �0.38 0.09
MG 4.34 1.55
NMSE 3.40 2.24
FSD 1.60 1.30
VG 203.6 53.1
r �0.02 0.19
FM 0.11 0.29
Statistical indicators assess the quality of a simulation based on
the ECMWF ERA-40 and operational meteorological fields. Due
to large discrepancies between model predictions and data
geometric mean bias (MG) and geometric mean variance (VG)
have been additionally reported. The indicators have been
computed for 118 measurements. The only exception is VG,
which has been computed for 108 measurements. The measure-
ments are those for which model prediction is higher than
1mBqm�3.
D. Quelo et al. / Atmospheric Environment 41 (2007) 5300–5315 5313
are also in progress. Sensitivity analysis arecurrently performed (following Brandt et al. (2002)for scavenging parameterizations and Mallet andSportisse (2005) for a comprehensive set of modelconfigurations). A parameterized version describingaerosols to which radionuclides are bound is alsodeveloped. The issue is to have an improveddescription of the scavenging processes. Someworks are also devoted to the opportunity of usingdata assimilation to improve forecast of radio-nuclides. Others are dedicated to the re-analysis offormer events, for instance the reconstruction of theChernobyl source term (Davoine and Bocquet,2007).
Acknowledgements
We thank our colleague Jean-Pierre Issartel forhaving initiated the ETEX runs at CEREA. Wethank Stefano Galmarini (JRC, Ispra) for havingprovided the observational data of Algeciras re-lease.
Appendix. Statistical indicators
The statistical indicators considered to be themost effective for the evaluation of the model-to-data comparison are defined in this section follow-ing Mosca et al. (1998) and Brandt (1998).
We considered a set of measurements fMigNi¼1
(mean M) and the corresponding model outputsfPig
Ni¼1 (mean P).
�
The bias (B) is defined as the average differencebetween the two sets of valueB ¼1
N
XN
i¼1
ðPi �MiÞ ¼ P�M. (A.1)
Its sign indicates an under or over prediction ofthe model with respect to the measurements.
� The fractional bias (FB) represents a relativedifference between model outputs and measure-ments (comprise in [�2,2]).
FB ¼ 2P�M
PþM. (A.2)
�
The geometric mean bias (MG) is meaningfulwhen the ratio between measurements and modeloutputs is largeMG ¼ exp1
N
XN
i¼1
ln Pi �1
N
XN
i¼1
ln Mi
!
¼YNi¼1
Pi
Mi
� �1=N
. ðA:3Þ
�
The normalized mean square error (NMSE)gives information on the deviations. It is definedasNMSE ¼1
N
XN
i¼1
ðPi �MiÞ2
PM(A.4)
�
The geometric mean variance (VG) gives infor-mation on the deviations as well. It is defined asVG ¼ exp1
N
XN
i¼1
ðln Pi � ln MiÞ2
" #. (A.5)
�
The fractional standard deviation (FSD) isdefined asFSD ¼ 2s2M � s2Ps2M þ s2P
, (A.6)
where
s2P ¼1
N
XN
i¼1
ðPi � PÞ2 and
ARTICLE IN PRESSD. Quelo et al. / Atmospheric Environment 41 (2007) 5300–53155314
s2M ¼1
N
XN
i¼1
ðMi �MÞ2: ðA:7Þ
�
The correlation coefficient ranges between �1and +1. It is defined ascorrelation ¼
PNi¼1ðMi �MÞðPi � PÞffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiPN
i¼1ðMi �MÞ2PNi¼1
ðPi � PÞ2
s .
(A.8)
�
The figure of merit (FM) is defined asFM ¼
PNi¼1 minðPi;MiÞPNi¼1 maxðPi;MiÞ
. (A.9)
�
The figure of merit in space (FMS) is calculatedat a fixed time, for a fixed concentration level,called the ‘significant level.’ It is defined as thepercentage of overlap of the measured (AM) andpredicted (AP) areas above the significant level.The FMS is given by the ratio between theintersection of the areas, and their union. � The figure of merit in time (FMT) corresponds tothe (FM) calculated at a fixed location.
References
Baklanov, A., 1999. Parameterisation of the deposition processes
and radioactive decay: a review and some preliminary results
with the DERMA model. Technical Report 99-4, DMI.
Baklanov, B., Sorensen, J., 2001. Parameterisation of radio-
nuclides deposition in atmospheric long-range transport
modelling. Physics and Chemistry of the Earth (B) 26 (10),
787–799.
Bocquet, M., 2005a. Grid resolution dependence in the recon-
struction of an atmospheric tracer source. Nonlinear Pro-
cesses in Geophysics 12, 219–234.
Bocquet, M., 2005b. Reconstruction of an atmospheric tracer
source using the principle of maximum entropy II: applica-
tions. Quarterly Journal of the Royal Meteorological Society
131 (Part B(610)).
Boutahar, J., Lacour, S., Mallet, V., Quelo, D., Roustan, Y.,
Sportisse, B., 2004. Development and validation of a fully
modular platform for numerical modelling of air pollution:
Polair3D. International Journal of Environment and Pollu-
tion 22 (1–2).
Brandt, J., 1998. Modelling transport, dispersion and deposition
of passive tracers from accidental releases. Ph.D. Thesis,
NERI, Denmark.
Brandt, J., Bastrup-Birk, A., Christensen, J.H., T., M., Thykier-
Nielsen, S., Zlatev, Z., 1998. Testing the importance of
accurate meteorological input fields and parameterizations
in atmospheric transport modelling, using DREAM—valida-
tion against ETEX-1. Atmospheric Environment 32 (24),
4167–4186.
Brandt, J., Christensen, J., Frohn, L., 2002. Modelling transport
and deposition or caesium and iodine from the Chernobyl
accident using the DREAM model. Atmospheric Chemistry
and Physics 2, 397–417.
Buckley, R.L., 1999. Modeling atmospheric deposition from a
cesium release in Spain using a stochastic transport model.
Technical Report Westinghouse Savannah River Company,
USA.
Davoine, X., Bocquet, M., 2007. Inverse modelling-based
reconstruction of the Chernobyl source term available for
long-range transport. Atmospheric Chemistry and Physics 7,
1549–1564.
Galmarini, S., Bianconi, R., Addis, R., Andronopoulos, S.,
Astrup, P., Bartzis, J.C., Bellasio, R., Buckley, R., Champion,
H., Chino, M., et al., 2004a. Ensemble dispersion forecast-
ing—part II: application and evaluation. Atmospheric En-
vironment 38 (28), 4,619–4,632.
Galmarini, S., Bianconi, R., Klug, W., Mikkelsen, T., Addis, R.,
Andronopoulos, S., Astrup, P., Baklanov, A., Bartniki, J.,
Bartzis, J.C., et al., 2004b. Ensemble dispersion forecasting—
part I: concept, approach and indicators. Atmospheric
Environment 38 (28), 4,607–4,617.
Graziani, G., Galmarini, S., Mikkelsen, T., 2000. RTMOD: Real-
Time Model Evaluation. Technical Report, Risøe, Denmark.
URL: /http://www.risoe.dk/rispubl/VEA/ris-r-1174.htmS.
Haas, H., Memmesheimer, M., Geiss, H., Jacobs, H., Laube, M.,
Ebel, A., 1990. Simulation of the Chernobyl radioactive cloud
over Europe using the EURAD model. Atmospheric Envir-
onment 24A (3), 673–692.
Issartel, J.-P., 2003. Rebuilding sources of linear tracers after
atmospheric concentration measurements. Atmospheric
Chemistry and Physics (3), 3111–3125.
Issartel, J.P., Baverel, J., 2003. Inverse transport for the
verification of the comprehensive nuclear test ban treaty.
Atmospheric Chemistry and Physics 3, 475–486.
JRC, 1998. ETEX, the European Tracer EXperiment: compre-
hensive overview of the ETEX project. Technical Report,
Joint Research Center.
Klug, W., Graziani, G., Gripa, G., Pierce, D., Tassone, C., 1992.
Evaluation of long range atmospheric transport models using
environmental radioactivity data from the Chernobyl acci-
dent: the ATMES report. Technical Report. Elsevier Applied
Science.
Krysta, M., Bocquet, M., Isnard, O., Issartel, J.P., Sportisse, B.,
2004. Data assimilation of radionuclides at short and regional
scales. In: Proceedings of the Workshop on Environmental
security. NATO, Kluiwer, Dordrecht.
Louis, J., 1979. A parametric model of vertical eddy fluxes in the
atmosphere. Boundary-Layer Meteorology 17, 197–202.
Mallet, V., Sportisse, B., 2005. A comprehensive study of ozone
sensitivity with respect to emissions over Europe with a
chemistry-transport model. Journal of Geophysical Research
110 (D22).
Mallet, V., Sportisse, B., 2006a. Ensemble-based air quality
forecasts: a multi model approach applied to ozone. Journal
of Geophysical Research 111 (D18), 18302.
Mallet, V., Sportisse, B., 2006b. Uncertainty in a chemistry-
transport model due to physical parameterizations and
numerical approximations: an ensemble approach applied to
ARTICLE IN PRESSD. Quelo et al. / Atmospheric Environment 41 (2007) 5300–5315 5315
ozone modeling. Journal of Geophysical Research 111
(D01302).
Mallet, V., Pourchet, A., Quelo, D., Sportisse, B., Investigation
of some numerical issues in Chemistry-Transport Models:
gas-phase simulations. Journal of Geophysical Research. In
revision.
Maryon, R., Smith, F., Conwy, B., Goddard, D., 1992. The UK
nuclear accident response model (NAME). Progress in
Nuclear Energy 26 (2), 85–104.
Mosca, S., Graziani, G., Klug, W., Bellasio, R., Bianconi, R.,
1998. A statistical methodology for evaluation of ling-range
dispersion models: an application to the ETEX exercise.
Atmospheric Environment 41 (24), 4307–4324.
Nodop, K. (Ed.), May 1997. ETEX Symposium on Long-range
Atmospheric Transport, Model Verification and Emergency
Response. Vienna, Austria.
Pasler-Sauer, J., 2003. Description of the atmospheric dispersion
model AT-STEP Version RODOS PV 5.0. Technical Report
RODOS(RA2)-TN(03)-01, Forschungszentrum Karlsruhe
GMBH.
Pollanen, R., Valkama, I., Toivonen, H., 1997. Transport of
radioactive particles from the Chernobyl accident. Atmo-
spheric Environment 31 (21), 3575–3590.
Quelo, D., Mallet, V., Sportisse, B., 2005a. Inverse modeling of
NOx emissions at regional scale over Northern France.
Preliminary investigation of the second-order sensitivity.
Journal of Geophysical Research 110 (D24310).
Quelo, D., Sportisse, B., Isnard, O., 2005b. Data assimilation for
short range dispersion of radionuclides: a case study for
second-order sensitivity. Journal of Environmental Radio-
activity 84, 393–408.
Robertson, L., Langner, J., 1997. Variational assimilation of
ETEX-I data. In: Nodop, K. (Ed.), Proceedings of the ETEX
Symposium on Long-range Atmospheric Transport, Model
Verification and Emergency Response.
Robertson, L., Langner, J., 1998. Source function estimate by
means of adjoint variational data assimilation applied to the
ETEX-I- tracer experiment. Atmospheric Environment 32
(24), 4219–4225.
Robertson, L., Perrson, C., 1993. Attempts to apply four
dimensional data assimilation of radiological data using the
adjoint technique. Radiation Protection Dosimetry 50 (2–4),
333–337.
Roustan, Y., Bocquet, M., 2006a. Inverse modeling for mercury
over Europe. Atmospheric Chemistry and Physics 6,
3,085–3,098.
Roustan, Y., Bocquet, M., 2006b. Sensitivity analysis for mercury
over Europe. Journal of Geophysical Research 111 (D14304).
Sportisse, B., 2007. A review of parameterizations for modeling
dry deposition and scavenging of radionuclides. Atmospheric
Environment 41, 2683–2698.
Thykier-Nielsen, S., Deme, S., Mikkelsen, T., 1999. Description
of the atmospheric dispersion module RIMPUFF. Technical
Report RODOS(WG2)-TN(98)-02, Forschungszentrum
Karlsruhe GMBH.
UNSCEAR, 2000. Annex J: exposure and effects of the
Chernobyl accident.
Verwer, J., Hundsdorfer, W., Blom, J., 26–29 October 1998.
Numerical time integration for air pollution models. In:
Proceedings of the Conference APMS’98. ENPC-INRIA.
Vogt, P., Pognaz, B., Aluzzi, F., Baskett, R., Sullivan, T., 1998.
ARAC Simulation of the Algeciras, Spain Steel Mill Cs-137
Release. Technical Report JC-131330 Lawrence Livermore
National Laboratory, USA.
