Atmospheric dispersion and ground deposition induced by the Fukushima Nuclear Power Plant accident: A local-scale simulation and sensitivity study

Abstract

Following the Fukushima Daiichi Nuclear Power Plant (FNPP1) accident on March 2011, radioactive products were released in the atmosphere. Simulations at local scale (within 80 km of FNPP1) were carried out by the Institute of Radiation Protection and Nuclear Safety (IRSN) with the Gaussian Puff model pX, during the crisis and since then, to assess the radiological and environmental consequences. The evolution of atmospheric and ground activity simulated at local scale is presented with a “reference” simulation, whose performance is assessed through comparisons with environmental monitoring data (gamma dose rate and deposition). The results are within a factor of 2–5 of the observations for gamma dose rates (0.52 and 0.85 for FAC2 and FAC5), and 5–10 for deposition (0.31 for FAC2, 0.73 for FAC5 and 0.90 for FAC10). A sensitivity analysis is also made to highlight the most sensitive parameters. A source term comparison is made between IRSN's estimation, and those from Katata et al. (2012) and Stohl et al. (2011). Results are quite sensitive to the source term, but also to wind direction and dispersion parameters. Dry deposition budget is more sensitive than wet deposition. Gamma dose rates are more sensitive than deposition, in particular peak values.

Full text

Atmospheric dispersion and ground deposition induced by the
Fukushima Nuclear Power Plant accident: A local-scale simulation
and sensitivity study
I. Korsakissok
*
, A. Mathieu, D. Didier
Institut de Radioprotection et de Sûreté Nucléaire (IRSN), PRP-CRI, SESUC, BMTA, Fontenay-aux-Roses 92262, France
highlights
<We model atmospheric dispersion of the Fukushima accident at local scale (80 km).
<Comparisons to gamma dose rate measurements are within a factor 2e5.
<Comparisons to deposition measurements are within a factor 5e10.
<Source term is the most sensitive parameter.
<Gamma dose rate, especially peak values, are more sensitive than deposition.
article info
Article history:
Received 15 October 2012
Received in revised form
21 December 2012
Accepted 3 January 2013
Keywords:
Atmospheric dispersion
Radionuclides
Fukushima
Model evaluation
Gamma dose rate
Sensitivity
abstract
Following the Fukushima Daiichi Nuclear Power Plant (FNPP1) accident on March 2011, radioactive
products were released in the atmosphere. Simulations at local scale (within 80 km of FNPP1) were
carried out by the Institute of Radiation Protection and Nuclear Safety (IRSN) with the Gaussian Puff
model pX, during the crisis and since then, to assess the radiological and environmental consequences.
The evolution of atmospheric and ground activity simulated at local scale is presented with a “reference”
simulation, whose performance is assessed through comparisons with environmental monitoring data
(gamma dose rate and deposition). The results are within a factor of 2e5 of the observations for gamma
dose rates (0.52 and 0.85 for FAC2 and FAC5), and 5e10 for deposition (0.31 for FAC2, 0.73 for FAC5 and
0.90 for FAC10). A sensitivity analysis is also made to highlight the most sensitive parameters. A source
term comparison is made between IRSN’s estimation, and those from Katata et al. (2012) and Stohl et al.
(2011). Results are quite sensitive to the source term, but also to wind direction and dispersion pa-
rameters. Dry deposition budget is more sensitive than wet deposition. Gamma dose rates are more
sensitive than deposition, in particular peak values.
Ó2013 Elsevier Ltd. All rights reserved.
1. Introduction
On March 11th 2011, an earthquake of magnitude 9.0 occur-
red off northeastern Japan, causing a tsunami and damaging the
Fukushima Daiichi Nuclear Power Plant (FNPP1). As a result,
radioactive products were released in the atmosphere. During
the emergency phase, the Institute of Radiation Protection and
Nuclear Safety (IRSN) was asked to provide its expertise in
support of the French authorities. Since then, the institute has
been working on improving its assessment of the terrestrial and
marine contamination (Mathieu et al., 2012;Bailly du Bois et al.,
2012). Understanding the formation process of highly con-
taminated areas cannot be achieved through measurements
only. While several kinds of measurements are available, they
only yield partial information: gamma dose rates devices have
a high temporal resolution, but are integrated overall gamma-
emitters, and are too scarce to provide a good spatial coverage.
Soil samplings and airborne readings provide maps of the con-
tamination, but no information on short-lived species, noble
gases, and temporal variations. Thus, improving atmospheric
dispersion simulations remains a key issue, especially for dose
assessments.
*Corresponding author. Tel.: þ33 1 58 35 85 49; fax: þ33146543989.
E-mail address: irene.korsakissok@irsn.fr (I. Korsakissok).
Contents lists available at SciVerse ScienceDirect
Atmospheric Environment
journal homepage: www.elsevier.com/locate/atmosenv
1352-2310/$ esee front matter Ó2013 Elsevier Ltd. All rights reserved.
http://dx.doi.org/10.1016/j.atmosenv.2013.01.002
Atmospheric Environment 70 (2013) 267e279

To this day, most numerical studies have focused on recon-
structing the source term using environmental data, either at large
scale (Stohl et al., 2011,Winiarek et al., 2012), or at local scale
(Katata et al., 2012), and using it for dose assessments (e.g. Morino
et al., 2011;Terada et al., 2012). Numerical simulations and model-
to-data comparisons at local scale are scarce, partly because of the
difficulty to produce satisfactory meteorological fields at that scale.
Such studies (Chino et al., 2011 and following papers) use gamma
dose rates, but not deposition measurements (except Terada et al.,
2012 which is based on prefectural measurements at Japan scale).
The existing studies were conducted with a given source term and
set of deposition parameters, but no extensive sensitivity study has
been carried out.
This paper presents the evolution of atmospheric and ground
activity simulated at local scale (within 80 km of FNPP1). Simu-
lations are made with pX, IRSN’s Gaussian puff model. The pX
model is part of the operational platform C3X, which is used by
IRSN’s Emergency Response Center in case of an accidental radi-
oactive release. The aim of this study is to better understand the
formation processes of the contaminated areas, but also to give
new insights on the pertinence and limitations of model evalua-
tion tools and indicators in accidental situations. Indeed, usual
Gaussian model evaluations are made on simple, well-known
dispersion experiments. The Fukushima accident provides an un-
precedented case to evaluate atmospheric dispersion models
devoted to radionuclides, with many environmental measure-
ments. We try to highlight advantages and shortcomings of each
kind of measurements and of statistical indicators used to evaluate
a model’s performance.
More than one year later, many uncertainties remain, especially
on the source term (release kinetics, source height, and isotopic
composition) and meteorology. Besides, uncertainties in simulation
parameters such as dry deposition velocities and scavenging co-
efficients cannot be neglected. Our sensitivity study is aimed at
identifying the most sensitive simulation parameters and input
data.
This paper is organized as follows: first, input data and simu-
lation set-up are described for a “reference”configuration (Section
2). Then, this reference simulation is compared to gamma dose rate
and deposition measurements (Section 3). Finally, the sensitivity
simulations are presented and discussed based on total deposition
budget and model-to-data comparisons (Section 4).
2. “Reference”simulation set-up and input data
2.1. Meteorological data
2.1.1. Wind
Operational forecasts from the European Center for Medium-
Range Weather Forecasts (ECMWF) with a spatial resolution of
0.125

0.125

and a time resolution of 3 h were used. The model
does not resolve the complex topography of the Fukushima area,
which is located close to the sea and within 10 km of a mountainous
area. The dataset was therefore analyzed and compared with
available meteorological data at several monitoring stations in the
Fukushima prefecture.
Fig. 1 shows the comparison between the wind at FNPP1 fore-
cast by ECMWF model, and observed by a monitoring car. The
modeled wind speed is often higher than the observed values,
which is consistent with the difference in heights between obser-
vations (monitoring car) and simulation (10 m). Besides, the cell
containing FNPP1 is partly over the ocean, where wind speeds are
globally higher. The wind direction comparison shows a rather
good model-to-data agreement except for some events, especially
during March 15 in the evening. During this period, observations
clearly indicate a northenorthwest plume travel direction, whereas
the modeled direction is mostly west. Accuracy in the wind direc-
tion is an issue of prime importance at that time since it coincides
with heavy rainfalls (cf. Section 2.1.2) and large releases. Therefore,
the simulations used in this study were carried out using homo-
geneous wind fields built with observations at FNPP1 during March
15, between 18 h and midnight, with a 10-min frequency. This so-
lution had its own limitations, since the wind observed at Daiichi
was used on a larger domain than its representativity scale, and did
not take into account vertical wind shear. For the rest of the sim-
ulation, three-dimensional ECMWF data were preferred, to account
for the heterogeneity of the flow.
Fig. 1. Comparison of wind observations (1-h median) at FNPP1 and wind given by ECMWF model, between March 12th and 20th. Left: wind direction (the bands of color indicate
the plume travel direction), right: wind speed. The ECMWF wind is taken in the cell of FNPP1 (without spatial interpolation), at 10 m. (For interpretation of the references to color in
this figure legend, the reader is referred to the web version of this article.)
I. Korsakissok et al. / Atmospheric Environment 70 (2013) 267e279268

2.1.2. Rain
During the accident, several rain episodes occurred. The first
one took place on 15e16 March and contributed to a significant
contamination by wet deposition. The second one happened on
20e23 March, both in the Fukushima region and in Tokyo. The
spatial and temporal resolution of the ECMWF model may be
insufficient to accurately represent the spatial and temporal
variability of rain episodes. On the other hand, rain radar obser-
vations provided by the Japan Meteorology Agency
1
have
a higher resolution in time and space but may oversee light and
localized rainfalls. Both datasets were compared to observed
rainfall rates available at several meteorological stations (not
shown). Fig. 2 illustrates that radar data improve the spatial
resolution of rain, even when averaged on the same grid reso-
lution as modeled data. It also shows the tendency of meteoro-
logical forecasts to overestimate the spatial coverage of rainfalls,
due to the coarse resolution. Thus, rain radar data are used in the
reference simulation.
2.2. Source term
The source term used here reconstructs atmospheric releases
occurring between March 12th and 26th, 2011. It includes 73 ra-
dioisotopes, based on the French reactor core inventory, corrected
to account for the nominal reactor power. The methodology used
for this estimation is detailed in other references (Corbin and Denis,
2012;Mathieu et al., 2012;IRSN, 2012).
The main release periods are given in Tab le 1. Two main
release heights were considered. A 20-m height (reactor pressure
vessel) is taken whenever pressure decreases happen in the unit,
and a 120-m height (exhaust stack) is preferred for specific
venting actions. For the hydrogen explosions (events 1 and 3), the
source was assumed to be diluted by the heat and momentum of
the explosion, within 100 m and 300 m on the vertical respec-
tively (Katata et al., 2012). After March 17th (events 7e10),
releases were detected by on-site monitoring devices, but could
not be associated to specific units or events. For these releases,
simulation results were used to infer the source height. For event
9, a height of 50 m was preferred, which is consistent with the
smokes coming from the top of reactor buildings (58 m) at that
time.
2
The radionuclides released in the atmosphere can be classified
into (1) noble gases that neither react with other species, nor de-
posit on the ground, and (2) species that undergo dry and wet
deposition. Fig. 3 shows that noble gases were preponderant
especially in the first releases (events 1, 2 and 3), since they are
more volatile. The total estimated activity released in the atmo-
sphere was 7.18 10
18
Bq (Becquerel). Noble gases (xenon and
krypton) contributed to most of the released activity (91% of the
activity), iodine to 6%, and cesium to less than 1%. For the major
contributors to gamma dose rate, the estimated released quantity
was 1.97 10
17
Bq for
131
I, 1.68 10
17
Bq for
132
I, 1.08 10
17
Bq for
132
Te, 2.06 10
16
Bq for
137
Cs, 2.78 10
16
Bq for
134
Cs and
5.94 10
18
Bq for
133
Xe. The isotopic ratio
131
I/
137
Cs in the source
term was initially 12 for Unit 1 and about 9 for Units 2 and 3. These
initial values are consistent with the ratio of 10 taken by Chino et al.
(2011), based on concentration measurements in rain water and
vegetation at local scale.
2.3. Dispersion and deposition parameters
IRSN’s Gaussian puff model pX was used for the simulations
(Soulhac and Didier, 2008). It handles radioactive decay and decay
products with a comprehensive mechanism. Dispersion in the pX
model is based on a discrete representation of the atmosphere
using PasquilleTurner stability classes (Turner, 1969). In our sim-
ulations, the stability was determined using the temperature gra-
dient between 2 m and 100 m in the meteorological forecasts, for
each cell and time step. The Pasquill standard deviation laws
(Pasquill, 1961) were used in the reference configuration.
Dry deposition and wet scavenging are crucial processes to
model the atmospheric behavior of radionuclides, since the species
deposed on the ground still contribute to the gamma dose-rate
after the plume departure from the area. The deposition prop-
erties of radionuclides strongly depend on three aspects: the gas/
aerosols partitioning, the proportion of organic and inorganic forms
of iodine, and the aerosol size distribution (Sportisse, 2007).
Fig. 2. Maps of rainfall rate on March, 15th at 23 h JST. Left: rain simulated by the meteorological model eRight: radar observations averaged on the same mesh (0.125resolution).
1
http://agora.ex.nii.ac.jp/earthquake/201103-eastjapan/weather/data/radar-
20110311/.
2
http://www.tepco.co.jp/en/press/corp-com/release/11032312-e.html.
I. Korsakissok et al. / Atmospheric Environment 70 (2013) 267e279 269

Without knowledge of these features, simple models were chosen
over detailed microphysical modeling. An apparent deposition ve-
locity of 0.2 cm s
1
is taken for all particles, notably
137
Cs (Brandt
et al., 2002). For iodine, 2/3 of released iodine quantity is
assumed to be gaseous (molecular) and 1/3 in particulate form. The
deposition velocity of molecular iodine I
2
is higher than for par-
ticulate matter: 0.7 cm s
1
(Baklanov and Sørensen, 2001). The
deposition velocities used over sea water are lower than over lands,
namely 0.05 cm s
1
(Pryor et al., 1999). The scavenging coefficient is
given by Ls¼L
0
p
0
B
, with p
0
the rain intensity in mm h
1
. In the
reference simulation, L
0
¼510
5
hmm
1
s
1
(as in Terada et al.,
2012) and B¼1.
3. Reference results and discussion
In this section, simulations with the pX model and the reference
configuration are analyzed and compared with observations. Sta-
tistical indicators used for model-to-data comparisons are defined
in Appendix.
3.1. Gamma dose rates
The observations used in this section come from prefectural
monitoring devices,
3
along with the data provided by TEPCO (Tokyo
Electric Power Company) at Fukushima Nuclear Power Plant 2,
hereafter called Daini.
4
Stations with a low-quality signal (too much
missing data) were discarded. In total, eight monitoring stations
were used (Table 2 and Fig. 7). To compute the simulated gamma
dose rates, 135 radionuclides (including 73 emitted species plus
their decay products) are taken into account. Dose coefficients
(Eckerman and Ryman, 1993) are used to infer dose rates from each
species’volume and surface activities. Since the Gaussian model
gives an analytical formula of the concentration, the dose rates
were directly computed at the stations’locations and no simulation
grid was used.
3.1.1. Overall performance on stations
The overall model performance on stations is satisfactory: 52%
of data are within a factor of 2 of the observations (FAC2), and 85%
are within a factor of 5 (FAC5). The fractional bias is 0.44 which
indicates a trend to overestimation, the correlation is 0.72, and the
figure of merit in time (FMT) is 0.43. This compares well to typical
Gaussian models performance on dispersion experiments
(Korsakissok and Mallet, 2009), even though uncertainties on input
data are much higher.
The ambient gamma dose rate measured in air comes from two
contributions: the direct plume contribution (hereafter called
“plume-shine”) and the gamma-ray emitted by radionuclides
Fig. 3. Time evolution of atmospheric release rate from Fukushima Daiichi Nuclear Power Plant between 12 and 26 March, 2011, for (a) Noble gas and (b) Other species (cesium,
iodine, tellurium.).
Table 1
Release periods of the source term, events that caused the releases (when identified), source height and main plume travel direction.
Number Beginning date Ending date Event Main plume travel direction Source height
1 12/03 10 h 13/03 09 h Unit 1 ehydrogen explosion North, then east 20 m (diluted on 100 m)
2 13/03 08 h 13/03 13 h Unit 3 eventing East (Pacific Ocean) 120 m
3 14/03 05 h 14/03 15 h Unit 3 eventing, then hydrogen explosion East (Pacific Ocean) 150 m (diluted on 300 m)
4 15/03 00 h 15/03 04 h Unit 2 - venting South 120 m
5 15/03 06 h 15/03 21 h Unit 2 ebreach on the wet-well West, north-west, south 20 m
6 16/03 00 h 16/03 15 h Units 2 and/or 3 pressure decreases South 20 m
7 18/03 15 h 18/03 20 h Unit 3? North 120 m
8 20/03 15 h 21/03 04 h Units 2 and/or 3? South 120 m
9 21/03 15 h 23/03 00 h Units 2 and 3 (white and gray smokes) South-west 50 m
10 25/03 08 h 25/03 10 h Unit 2? West 120 m
3
http://www.pref.fukushima.jp/j/7houbu0311-0331.pdf and http://www.pref.
fukushima.jp/j/20-50km0315-0331.pdf.
4
http://www.tepco.co.jp/en/nu/fukushima-np/f2/data/2011/index-e.html.
I. Korsakissok et al. / Atmospheric Environment 70 (2013) 267e279270

deposed on the ground (“ground-shine”). The plume-shine usually
is responsible for peak values (high but during a short time),
whereas the ground-shine corresponds to the gamma dose-rate
measured after the plume departure. Hence, the most numerous
gamma-dose rate observations correspond to ground-shine, which
decreases slowly due to radioactive decay. The decrease rate of
this residual gamma dose rate depends on the radioisotopes
deposed on the ground, and on their respective half-life times.
Therefore, “classical”indicators such as FAC2 and FAC5 mainly
depend on the simulation’s ability to forecast deposition, isotopic
composition and subsequent decay. In case of an accidental release
of radionuclides, an operational simulation should be able to
forecast peak values, since they represent an important part of
human exposure (through direct radiation and inhalation), and
plume arrival times (i.e. the first date when the gamma dose rate
value is higher than background value). In the following, we will
focus on (1) bias on peak values, (2) plume arrival times and (3)
FAC2, FAC5 and FMT. Comparisons are made on hourly-averaged
values.
Table 2 gives an overview of the model’s performance for each
station. Peak values are within less than a factor of two, except for
Iwaki, Daini and Minamisoma, where it is overestimated by a fac-
tor of five. These three stations are located close to the coast,
where the meteorological model often has difficulties forecasting
the wind field. Besides, for the two events responsible for the
peaks (event 1 for Minamisoma and event 4 for Daini and Iwaki),
the meteorological conditions were very stable. Thus, the plume
was very thin, and uncertainties in the wind field, station location,
and/or release height would have a large impact on the result. At
Minamisoma and Iwaki, gamma dose-rate measurements are only
available every hour during the main releases periods (prior to
March 16). Simulations show a high temporal variability, especially
during the plume passage, indicating that the temporal frequency
of observations may not be sufficient. The temporal resolution of
the wind field (3 h) is also too coarse to account for the wind
variability.
On the other stations, the peak values are very well repro-
duced, although a delay of 6 h in the plume arrival time is
observed on the northwestern stations Iitate and Fukushima. At
Funehiki, there were no observations at the simulated peak
times, and the air dose-rate due to deposition is overestimated
by more than a factor of five. For all other stations except Iwaki,
more than 80% of simulated values are within a factor 5 of the
observations, and the FAC2 is also very good, especially at the
northwestern stations. Fig. 4 shows the model-to-data compari-
sons of gamma dose rate values, for the eight monitoring sta-
tions, hour-by-hour. A few singular values are underestimated,
especially at Iitate and Fukushima, due to the delay in the plume
arrival time.
3.1.2. Temporal analysis on stations
Fig. 5 shows the temporal evolution on four of these stations:
two are representative of a high contamination level due to wet
deposition (Iitate and Fukushima), and two stations where little
wet deposition occurs (Minamisoma and Kawauchi). The latter
show a time series with several peaks corresponding to plumes
coming through the station, and little deposition, while the former
show a high contamination by deposition, with a decrease rate
over time essentially due to radioactive decay of the deposed
isotopes.
At Minamisoma and Kawauchi (Fig. 5(a) and (b)), most peaks
are simulated, with a small delay and some overestimation at
Kawauchi. Missing peaks are probably due to inaccuracies in the
source term, but are small compared to the initial con-
tamination. These two stations illustrate the difficulty to cor-
rectly represent both peak values and deposition: at
Minamisoma, the peak value is overestimated but the deposition
is correctly forecast, while at Kawauchi, the peak value is much
better reproduced but the deposition is overestimated after
March 22nd. At Iitate and Fukushima, the dose rates are well
reproduced, except for the initial delay in the plume arrival (6 h).
This delay is also found by other simulations (Katata et al., 2012).
Fig. 6 illustrates the contamination processes at two typical
Table 2
Comparison of observed and simulated ambient air dose rate at the eight stations: peak time and value (the peak is the maximum value on the whole period of observations),
Fac2 and Fac5 (proportion of simulation values within a factor of 2 resp. 5 of the observations). Statistics are made on the whole period (12/03e26/03).
Station name Event(s) associated
with peaks
Observed arrival time
(1st peak)
Simulated arrival
time (1st peak)
Observed peak
value (
m
Gy h
1
)
Simulated peak
value (
m
Gy h
1
)
Fac2 (%) Fac5 (%) FMT
Kawauchi Event 5; event 9 15/03 11:00 15/03 12:00 11.5 15.7 69 99 0.64
Tamura City
Funehiki
Event 5 (2 peaks) e15/03 16:00
16/03 06:00
e6.2
9.4
5 82 0.19
Koriyama Event 5; event 9 15/03 14:00 15/03 17:00 6 3.0 52 99 0.47
Iitate Event 5 15/03 15:00 15/03 21:00 39.5 57.6 67 97 0.54
Fukushima Event 5 15/03 16:00 15/03 22:00 24 19.1 92 98 0.79
Minamisoma Event 1; event 7 12/03 20:00 12/03 19:00 20.0 87 95 100 0.68
Daini (FNPP2) Event 4; event 6 15/03 00:00 15/03 01:00 94 515 12 88 0.38
Iwaki Event 4; event 6 15/03 01:00 15/03 03:00 23.7 147 9 16 0.12
Fig. 4. Scatter plot of gamma dose rate values (
m
Gy h
1
) at the eight monitoring sta-
tions for the reference simulation. Red line: perfect agreement. Dashed line (bold):
factor 2. Dashed line: factor 5.
I. Korsakissok et al. / Atmospheric Environment 70 (2013) 267e279 271

stations: (a) Kawauchi, where the plume contribution is pre-
ponderant (91% of the peak gamma dose rate) and ground shine
is only due to dry deposition, and (b) Fukushima, where the wet
deposition contribution to the gamma air dose rate is predomi-
nant (95% of the peak dose rate). Fig. 6(b) shows that the delay at
Fukushima is not due to inaccuracies in the rain timing, but to
a delay in the plume arrival time. Indeed, the plume contribution
does not increase significantly before 22 h on March 15th. This
might be due to uncertainties in the wind field and/or source
height.
Fig. 5. Comparison of observed and simulated gamma air dose rate at four of the eight monitoring stations. Simulations are made with pX, for the reference configuration.
Fig. 6. Plume, dry deposition and wet deposition contributions to the simulated ambient air gamma dose rate at (a) Kawauchi and (b) Iitate.
I. Korsakissok et al. / Atmospheric Environment 70 (2013) 267e279272

3.2. Deposition
The simulations used in this section are made on a polar mesh,
with circles at 2.5, 5, 10, 15, 20, 25, 30, 40, 50, 65 and 80 km from
FNPP1, and a ten-degree step for the angle.
3.2.1. Spatial analysis
Fig. 7 shows the simulated spatial distribution of deposition
after the end of releases (March 30, 2011). Dry deposition mostly
occurred along the coast, north and south from FNPP1 ((Fig. 7(a)),
whereas wet deposition strongly contaminated a northwestern
area, including Iitate and Fukushima (Fig. 7(b)). Analyses by event
showed that the dry deposition north of FNPP1 was formed
during two distinct events: event 1 (12/03) and event 7 (18/03).
The southern deposition along the coast corresponds to event 4
(15/03 at 00 h) and event 6 (16/03). Wet deposition mainly cor-
responds to event 5 (67% of total land deposition occurred within
few hours on March 15th), but also occurred on 21/03 and 22/03
(16% of total deposition), south and south-west of the plant
(event 9). The dry deposition over the sea is much lower, due to
the use of lower deposition velocities over water. The con-
tribution of dry deposition to total deposition is shown Fig. 7(c).
It confirms that the northwestern contamination is mainly due to
wet deposition (more than 90% of the total deposition), whereas
dry deposition contributes to almost 100% deposition along the
coast.
Dry and wet deposition patterns are similar for other species
than
137
Cs, although dry deposition is stronger for molecular iodine.
This leads to a higher ratio
131
I/
137
Cs where dry deposition is pre-
dominant, especially south of FNPP1 (Fig. 7(d)). This pattern is
consistent with observations (Kinoshita et al., 2011). This ratio
tends to decrease over time, since the half-life of
131
I (eight days) is
much lower than for
137
Cs (thirty years).
3.2.2. Deposition budget
As Morino et al. (2011) did on Japan scale, we computed the total
budget of
137
Cs and
131
I deposed on the simulation domain through
dry and wet deposition (decay-corrected). Their source term comes
from Chino et al. (2011), which includes 13 PBq of
137
Cs (20.6 PBq in
our estimation) and 150 PBq of
131
I (197 PBq in our case). Morino
et al. (2011) supposed that the gaseous fraction of
131
I was 80% of
the total release (67% in our case). Their estimation for the
Fukushima prefecture can be compared with our land deposition
budget within 80 km from FNPP1 (Table 3).
The absolute values are of the same order of magnitude, except
for
137
Cs dry deposition, where our value is ten times higher. This
Fig. 7. Spatial distribution of (a)
137
Cs dry deposition, (b)
137
Cs wet deposition, (c) Ratio of dry deposition on total deposition for
137
Cs and (d)
131
I/
137
Cs ground activity ratio. Values
are given on March 30, 2011.
I. Korsakissok et al. / Atmospheric Environment 70 (2013) 267e279 273

may be due to the lower released amount, and to lower dry dep-
osition velocities (0.1 cm s
1
instead of 0.2 cm s
1
in our simula-
tions). For gaseous
131
I, Morino et al. (2011) took 0.5 cm s
1
(0.7 cm s
1
in our case). Estimates of total iodine dry deposition are
comparable. Wet deposition values are more difficult to compare,
without knowledge of the parameterization used. In our case,
scavenging is the same for all species. In Morino et al. (2011), the
wet deposition scheme seems more efficient for
137
Cs, which is in
particulate form, than for
131
I, which is mostly gaseous. It probably
explains the higher proportion of
137
Cs deposed over land (12% of
their release).
Total deposition values are very close to each other. Compen-
sation between dry and wet deposition partly explains this agree-
ment. Based on airborne monitoring data
5
, the total quantity of
137
Cs within 80 km of the plant over land was estimated to be about
1.49 PBq, with an uncertainty of 0.49 PBq (Gonze, M-A, personal
communication). Both simulations are within this range.
3.2.3. Comparisons to deposition measurements
Fig. 8 shows the comparison between the simulation and
the observations of
137
Cs deposition provided by Ministry of
Education, Culture, Sports, Science and Technology (MEXT).
6
About 1800 measurements are within our simulation domain.
The overall shape of the northwestern contamination (over
10
5
Bq m
2
) is correct, but the highest values are located too
north compared to the observations. This is probably due to the
use of wind observations at FNPP1 at that time, not representa-
tive of the wind direction at a larger scale. Thus, the model
overestimates deposition north along the coast by more than
a factor five (yellow to red points, Fig. 8(c)). The overestimation
south, after 40 km, is consistent with gamma dose rates at Iwaki.
Fig. 8(d) shows that the agreement is better between 20 and
60 km from the source. Closer to the source, the model tends to
overestimate deposition. Between 60 and 80 km, values are
underestimated, especially in the northern area (gray and purple
points, Fig. 8(c)). In all, 31% of simulated values are within a factor
2 of the observations, 73% within a factor 5, and 90% within
a factor 10. The correlation coefficient is 0.34. The figure of merit
in space (FMS) depends on the chosen threshold. A high
threshold would focus on the model’s ability to forecast extreme
values. For 10
4
Bq m
2
(94% of measurements), the FMS is very
good (0.85). For 10
5
Bq m
2
(30% of measurements, mainly
northwest), it goes down to 0.43, probably because of the mis-
placed wet deposition zone.
Very close measurement points may differ by a factor 5e10,
which may be due to running water or changes in terrain type
(school yard, agricultural field.). The simulation gives averaged
values, and does not account for local-scale variability. Thus, fur-
ther analyses of the datasets and aggregating close measurements
could improve these comparisons.
4. Sensitivity study
4.1. Sensitivity parameters
Table 4 shows an overview of the parameters chosen for the
sensitivity simulations. Simulations were carried out independ-
ently for each parameter, to assess their impact on results.
For dispersion, several Gaussian standard deviation formulations
were used. For dry and wet deposition, values were taken within the
range of the literature, i.e. more or less within a decade (Sportisse,
2007). Mixing the release between 0 and 150 m allowed high-
lighting the sensitivity to the release height. Several source terms
were also compared. The release from Katata et al. (2012) was
designed at local scale, including several isotopes to compute the
gamma dose rates. Thus, it was included in deposition and gamma
dose rate sensitivity runs. The release from Stohl et al. (2011), cali-
brated on long-range simulations (with a first-guess), did not
include short-lived species that are major contributors to gamma
dose rate. Thus, it was only used in the comparisons for
137
Cs dep-
osition. Another IRSN’s estimation, using inverse modeling with
gamma dose rate observations (Saunier et al., 2012), was included.
Concerning meteorological data, the influence of rain (radar obser-
vations versus rain forecasts) and wind fields (with and without
wind observations at Daiichi on March 15th) was evaluated.
4.2. Sensitivity results
4.2.1. Sensitivity of total deposition budget of
137
Cs
Fig. 9 illustratesthe sensitivity of the deposition budget overland.
The sensitivity of total deposition is conditioned by that of wet
deposition, whichaccounts for 2/3 of total deposition (cf. Table3). For
total deposition, most simulations are within a factor 2 of the refer-
ence simulation, except with the source term from Stohl et al. (2011)
which has a higher estimation of
137
Cs release (35 PBq of
137
Cs). Ac-
cording to airborne observations, the total land deposition within
80 km of FNPP1 should be between 1 and 2 PBq (see Section 3.2.2).
Most simulations meet this condition except the Release_Stohl and
lmin configurations. Dry deposition (Fig. 9(a)) is also sensitive to
vertical diffusion. Indeed, Briggs urban and constant diffusion pa-
rameterizations enhance vertical diffusion, which induces lower dry
deposition, since the plume is less concentrated near the ground.
Finally, a compensation mechanism between dry andwet deposition
appears: lower deposition velocities (vdmin) imply that the plume is
less depleted near the source. Hence, the plume wash-out by the rain
is more efficient, and wet deposition increases (Fig. 9(b)).
Table 3
Simulated
137
Cs and
131
I loss due to dry and wet deposition: contribution over land and sea, and total on the simulation domain (80 km around FNPP1). (*): deposed quantity
normalized by released quantity. Simulations are made with pX, for the reference configuration. Values are given in PBq (10
15
Bq). The values are decay-corrected for iodine.
Morino et al., 2011
Fukushima pref. 13,783 km
2
Land 80 km of FNPP1
about 10
4
km
2
Sea 80 km of FNPP1
about 10
4
km
2
Total (land þsea,
80 km radius)
131
I (PBq) Decay-
corrected
Dry deposition 8.07 7.9 0.3 8.2
Wet deposition 3.6 6.1 22.0 28.1
Total deposition 12.3 (8.2%)* 14.0 (7.1%)* 22.3 36.3 (18.4%)*
137
Cs (PBq) Dry deposition 0.047 0.39 0.02 0.39
Wet deposition 1.48 0.94 2.6 3.9
Total deposition 1.53 (11.7%)* 1.33 (6.4%)* 2.62 4.29 (20.8%)*
5
http://www.mext.go.jp/english/incident/1304796.htm.
6
http://www.mext.go.jp/b_menu/shingi/chousa/gijyutu/017/shiryo/__icsFiles/
afieldfile/2011/09/02/1310688_1.pdf.
I. Korsakissok et al. / Atmospheric Environment 70 (2013) 267e279274

4.2.2. Sensitivity of spatial deposition of
137
Cs
The sensitivity of spatial deposition may be investigated
through statistical indicators used in comparison to MEXT mea-
surements (Fig. 10). They are quite robust to changes in parame-
terizations such as standard deviations, vertical mixing, and dry
deposition. The source term is the most sensitive, along with rain
and wind fields. The Release_Stohl decreases significantly the FAC2,
and changes in meteorological data also reduce the FAC2
(Fig. 10(a)). A low scavenging coefficient (lmin) increases FAC2
(0.37), but lowers FAC5 (0.70). FAC2 and FAC5 would not penalize
a simulation that would homogenize the deposition. For instance,
a“simulated”homogeneous deposition of 10
5
Bq m
2
, compared to
Fig. 8. Comparison of observed and simulated deposition values for
137
Cs. (a) Observations from MEXT. (b) Map of bias factor (observation/simulation). (c) Deposition simulated
with pX. (d) Scatter plot: Red line: perfect agreement. Dashed line (bold): factor 2. Dashed line: factor 5.
Table 4
Sensitivity simulations: overview of all simulation parameters modified in the sensitivity runs. Each parameter is made to vary independently from one another. For each
sensitivity parameter, the name used in the figures is given in brackets.
Parameter Reference Value 1 (name) Value 2 (name) Value 3 (name)
Standard deviations Pasquill Briggs rural (briggs_rural) Briggs urban (briggs_urban) Constant diffusion K (diff)
Dry deposition particles
(iodine)
210
3
(7 10
3
)ms
1
510
4
(1 10
3
)ms
1
(vdmin)510
3
(2 10
2
)ms
1
(vdmax)
Wet deposition 5 10
5
hmm
1
s
1
110
5
(lmin)110
4
(lmax)
Release height Time varying (cf Table 1) Diluted between 0 and 150 m
(Vertical_mixing)
Source term Mathieu et al., 2012 Katata et al., 2012 (Release_Chino)Stohl et al., 2011 (Release_Stohl)Saunier et al., 2012
(Release_Inverse)
Rain Rain radar observations ECMWF forecasts (ECMWF_rain)
Wind ECMWF þobservations
at FNPP1
ECMWF forecasts (ECMWF_wind)
I. Korsakissok et al. / Atmospheric Environment 70 (2013) 267e279 275

observations, would have 0.33 as FAC2 and 0.73 as FAC5, which is
within the other simulations’results. Thus, other indicators such as
FMS (Fig. 10(b)) and deposition maps (Supplementary material)are
used as a complement.
The FMS is compared for a 10
5
Bq m
2
threshold which cor-
responds to the northwestern deposition. It is therefore sensitive
to wet deposition parameters (lmax,lmin) and to the wind used
during March 15th (ECMWF_wind). The Release_Stohl tends to
overestimate deposition in the northwestern area, but is not
penalized by this indicator, and the Release_Chino improves the
FMS.
4.2.3. Sensitivity of gamma dose rates
Gamma dose rates are much more sensitive than deposition,
which is integrated in time and has a better spatial coverage.
With only eight stations, local discrepancies between model and
reality on one station may have a huge impact on the results.
Thus, a simulation may have acceptable results on deposition,
but not on gamma dose rates. This is the case of briggs_urban
vertical_mixing and Release_inverse simulations (Fig. 11). Lower
dry deposition velocities improve the FMT by compensating for
the overestimation due to errors in the source term and/or
meteorological data. Dispersion parameters and source height
have a much larger influence on gamma dose rates, especially
peak values, than on deposition. This is particularly true on
coastal stations, where the plume is very thin and the peak
values are very sensitive to changes in the wind direction and
diffusion. For instance, using a constant diffusion coefficient
(diff) improves the results on coastal stations by increasing the
plume vertical dilution in stable situations (Fig. 12(a) and (c)),
thus reducing the overestimation. Using ECMWF forecasts
without observations at FNPP1 (ECMWF_wind) drastically lowers
the model performance on stations because of errors in the
westerly wind direction on March 15th: Iitate and Fukushima are
missed by the plume, and the peak at Koriyama is greatly
overestimated (Fig. 12(b) and (d)). Fig. 12 illustrates the difficulty
Fig. 9. Sensitivity of total
137
Cs deposition budget over land, within 80 km of FNPP1. Values for the reference simulation (similar to Table 3) are given in red (dashed lines). (For
interpretation of the references to color in this figure legend, the reader is referred to the web version of this article.)
I. Korsakissok et al. / Atmospheric Environment 70 (2013) 267e279276

to have a good model-to-data agreement (i.e. less than a factor 5)
on all stations: the Release_inverse simulation greatly improves
the results on coastal station but overestimates the northwest-
ern peaks, while the Release_chino is better on northwestern
stations but underestimates the peaks on southern and western
stations.
5. Conclusions and perspectives
We presented atmospheric dispersion simulations of the
Fukushima Nuclear Power Plant accident, using IRSN’s Gaussian
puff model pX. The evolution of atmospheric and ground activity
simulated at local scale was presented with a “reference”simu-
lation, whose performance was assessed through comparisons with
environmental monitoring data (gamma dose rate and deposition).
The results were within a factor of 2e5 of the observations for
gamma dose rates, and 5e10 for deposition. The total quantity of
137
Cs deposed over land and the isotopic ratio
131
I/
137
Cs were
consistent with observations and with other estimations. The
gamma dose rates in the northwestern area, highly contaminated
by wet deposition, were correctly reproduced but the wet
deposition zone was slightly misplaced. The coastal stations were
more difficult to simulate, due to the stable situation inducing
a thin plume, hence a large dependency to uncertainties in the
wind direction and station location. Besides, the temporal
frequency of measurements and/or wind fields may not be
sufficient to correctly reproduce the peak values when the plume
passed through the stations in less than 1 h. Thus, it is difficult to
determine whether the overestimation along the coast is due to
the source term, meteorological data and/or diffusion parameters.
Fig. 10. Sensitivity of comparisons to MEXT measurements of
137
Cs, within 80 km of FNPP1. Values for the reference simulation are given in red (dashed line). (For interpretation of
the references to color in this figure legend, the reader is referred to the web version of this article.)
Fig. 11. Sensitivity of comparisons to gamma dose rate measurements on the eight monitoring stations. Values for the reference simulation are given in red (dashed line). (For
interpretation of the references to color in this figure legend, the reader is referred to the web version of this article.)
I. Korsakissok et al. / Atmospheric Environment 70 (2013) 267e279 277

While uncertainties in the input data (wind direction, rain,
source term kinetics and quantity) are huge, simulation parameters
are also uncertain. The influence of each parameter was assessed
separately. As expected, source terms had the highest impact on the
results: the release from Stohl et al. (2011) tended to over-
estimation; that from Katata et al. (2012) gave better results on the
northwest but underestimation in the south and west. Total dep-
osition budget was within a factor 2 of the reference most of the
time. The sensitivity study seemed to validate our choice for scav-
enging coefficients. Gamma dose rates, especially peak values, were
quite sensitive: source terms, standard deviations, and wind di-
rection had huge impacts on the results. However, with only eight
monitoring stations, each station has a particular behavior and no
general conclusions can be made.
This study is a first step toward updating model evaluation
tools and proposing new indicators for modeling accidental re-
leases of radionuclides. Despite the situation complexity and the
remaining uncertainties, the results presented here are satisfactory
compared to standard dispersion model evaluations. They can
therefore be used as a benchmark for atmospheric dispersion
models’evaluation and improvement with respect to emergency
purposes. The main perspective is to use ensemble simulations
instead of a single, deterministic model, in order to account for
uncertainties in the input data and simulation parameters (Mallet
and Sportisse, 2008). Another idea is to account for uncertainties
in the plume position by comparing measurements to a cloud of
points, or to an average over a given volume, instead of a single
value.
Acknowledgments
We want to express our sympathy to Japanese people who
endured the terrible consequences of the earthquake, tsunami and
nuclear accident.
We thank MEXT and TEPCO for the online publication of
measurements.
We thank Météo France for providing meteorological data, D.
Corbin and J. Denis for the source term estimation, R. Gurriaran for
Fig. 12. Sensitivity of gamma dose rate peak values on four monitoring stations. Values for the reference simulation are given in red (dashed line). Observation values are given in
black (bold dashed line). (For interpretation of the references to color in this figure legend, the reader is referred to the web version of this article.)
I. Korsakissok et al. / Atmospheric Environment 70 (2013) 267e279278

his expertise on measurements, and people from the Emergency
Response Center who worked during the crisis.
Appendix A. Statistical indicators
We consider a set of measurements O
i
(ibetween 1 and N) and
the corresponding model outputs P
i
. Statistical indicators are
defined as follows (Chang and Hanna, 2004):
(1) Fractional bias MFBE:
MFBE ¼X
N
i¼0
P
i
O
i
P
i
þO
i
(2) Bias Factor at a given point:
BF ¼P
i
O
i
(3) Correlation coefficient:
r¼P
N
i¼1
O
i
OP
i
P
ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
P
N
i¼1
O
i
O
2
P
N
i¼1
P
i
P
2
r
(4) FAC2 (resp. FAC5): proportion of values such that 0.5 BF 2
(resp. 0.2 BF 5).
(5) The Figure of Merit in Time (FMT) is the percentage of overlap
between measured and predicted integrated time series at
a given location:
FMT ¼P
N
i¼1
minðO
i
;P
i
Þ
P
N
i¼1
maxðO
i
;P
i
Þ
(6) The Figure of Merit in Space (FMS) is the percentage of overlap
between measured and predicted areas above a threshold Tat
a given time. A
P
is the number of points verifying P
i
Tand A
O
is the number of points such that O
i
T, then
FMS ¼A
P
XA
O
A
P
WA
O
A“perfect”model-to-data agreement corresponds to MFBE ¼0
and other indicators equal to 1.
Appendix B. Supplementary material
Supplementary material related to this article can be found at
http://dx.doi.org/10.1016/j.atmosenv.2013.01.002.
References
Bailly du Bois, P., Laguionie, P., Boust, D., Korsakissok, I., Didier, D., 2012. Estimation
of marine source-term following Fukushima Daiichi accident. Journal of Envi-
ronmental Radioactivity 114, 2e9.
Baklanov, A., Sørensen, J.H., 2001. Parameterization of radionuclide deposition in
atmospheric long-range transport modelling. Physics and Chemistry of the
Earth (B) 26, 787e799.
Brandt, J., Christensen, J.H., Frohn, L., 2002. Modelling transport and deposition of
caesium and iodine from the Chernobyl accident using the DREAM model.
Atmospheric Chemistry and Physics 2, 397e417.
Chang, J.C., Hanna, S.R., 2004. Air quality model performance evaluation. Meteor-
ology and Atmospheric Physics 87, 167e196.
Chino, M., Nakayama, H., Nagai, H., Terada, H., Katata, G., Yamazawa, H., 2011.
Preliminary estimation of release amounts of
131
I and
137
Cs accidentally dis-
charged from the Fukushima Daiichi Nuclear Power Plant into the atmosphere.
Journal of Nuclear Science and Technology 48, 1129e1134.
Corbin, D., Denis, J., 2012. Evaluation des rejets atmosphériques liés à l’accident de
Fukushima. IRSN. Report PSN-RES/SAG/2012e00347.
Eckerman, K.F., Ryman, J.C., 1993. Exposure-to-dose Coefficients for General
Application, Based on the 1987 Federal Radiation Protection Guidance. Oak
Ridge National Laboratory. Report 12.
IRSN, 2012. Fukushima, One Year Later: Initial Analyses of the Accident and
Its Consequences. IRSN. Report IRSN/DG/2012-003. http://www.irsn.fr/EN/
publications/technical-publications/Documents/IRSN_Fukushima-1-year-later_
2012-003.pdf.
Katata, G., Ota, M., Terada, H., Chino, M., Nagai, H., 2012. Atmospheric discharge
and dispersion of radionuclides during the Fukushima Daiichi Nuclear Power
Plant accident. Part I: source term estimation and local-scale atmospheric
dispersion in early phase of the accident. Journal of Environmental Radio-
activity 109, 103e113.
Kinoshita, N., Sueki, K., Sasa, K., Kitagawa, J.-i., Ikarashi, S., Nishimura, T., Wong, Y.-S.,
Satou, Y., Handa, K., Takahashi, T., Sato, M., Yamagata, T., 2011. Assessment of
individual radionuclide distributions from the Fukushima nuclear accident
covering central-east Japan. Proceedings of the National Academy of Sciences
108 (49), 19526e19529.
Korsakissok, I., Mallet, V., 2009. Comparative study of Gaussian dispersion formulas
within the Polyphemus platform: evaluation with Prairie Grass and Kincaid
experiments. Journal of Applied Meteorology 48, 2459e2473.
Mallet, V., Sportisse, B., 2008. Air quality modeling: from deterministic to stochastic
approaches. Computers & Mathematics with Applications 55, 2329e2337.
Mathieu, A., Korsakissok, I., Quélo, D., Groëll, J., Tombette, M., Didier, D., Quentric, E.,
Saunier, O., Benoit, J.-P., Isnard, O., 2012. Atmospheric dispersion and deposition
of radionuclides from the Fukushima Daiichi nuclear power plant accident.
Elements 8, 195e200.
Morino, Y., Ohara, T., Nishizawa, M., 2011. Atmospheric behavior, deposition, and
budget of radioactive materials from the Fukushima Daiichi nuclear power
plant in March 2011. Geophysical Research Letters 38, L00G11.
Pasquill, F.A., 1961. The estimation of the dispersion of windborne material. The
Meteorological Magazine 90 (1063), 33e49.
Pryor, S.C., Barthelmie, R.J., Geernaert, L.L.S., Ellermann, T., Perry, K.D., 1999.
Speciated particle dry deposition to the sea surface: results from ASEPS ’97.
Atmospheric Environment 33, 2045e2058.
Saunier, O., Mathieu, A., Didier, D., Tombette, M., Quélo, D., Winiarek, V.,
Bocquet, M., 2012. Using gamma dose rate monitoring with inverse modelling
techniques to estimate the atmospheric release of a nuclear power plant acci-
dent: application to the Fukushima case. In: International Meeting on Severe
Accident Assessment and Management: Lessons Learned from Fukushima Dai-
ichi., San Diego, USA, 11e15 November 2012.
Soulhac, L., Didier, D., 2008. Projet pX, note de principe pX 1.0. IRSN. Report IRSN/
DEI/SESUC/08-39.
Sportisse, B., 2007. A review of parameterization for modelling dry deposition and
scavenging of radionuclides. Atmospheric Environment 41, 2683e2696.
Stohl, A., Seibert, P., Wotawa, G., Arnold, D., Burkhart, J.F., Eckhardt, S., Tapia, C.,
Vargas, A., Yasunari, T.J., 2011. Xenon-133 and caesium-137 releases into the
atmosphere from the Fukushima Dai-ichi nuclear power plant: determination
of the source term, atmospheric dispersion, and deposition. Atmospheric
Chemistry and Physics Discussions 11, 28319e28394.
Terada, H., Katata, G., Chino, M., Nagai, H., 2012. Atmospheric discharge and dis-
persion of radionuclides during the Fukushima Dai-ichi Nuclear Power Plant
accident. Part II: verification of the source term and analysis of regional-scale
atmospheric dispersion. Journal of Environmental Radioactivity 112, 141e154.
Turner, D.B., 1969. Workbook of Atmospheric Dispersion Estimates.
Winiarek, V., Bocquet, M., Saunier, O., Mathieu, A., 2012. Estimation of errors in the
inverse modeling of accidental release of atmospheric pollutant: application to
the reconstruction of the cesium-137 and iodine-131 source terms from the
Fukushima Daiichi power plant. Journal of Geophysical Research 117, D05122.
I. Korsakissok et al. / Atmospheric Environment 70 (2013) 267e279 279