advanced meteorological pre-processing for the real-time emergency response systems dealing with the...
TRANSCRIPT
Advanced meteorological pre-processing for the real-time emergency response systems dealing
with the atmospheric dispersion in complex terrain
I. Kovalets (IPMMS NAS of Ukraine), S. Andronopolous (NCSR “Demokritos”, Greece),
J. Bartzis (Thessaloniki University, Greece)
The situation
Real-Time On-Line Decision Support System for Nuclear Emergency Management
In Europe (RODOS)
AtmosphericDispersion Model
(ADM)
MeteorologicalPre-processor
(MPP)
Other Modules
…
Measurement data frommeteorological stations
NWP prognosticMeteorological data
•ADMs: Key Role in DSSs – determine the current, and predict the future spatial distribution of radionuclides after an accidental release of radioactivity to the atmosphere•MPPs: Interface between the ADMs and the incoming meteorological data•Meteorological data: measurements from one or more stations in the vicinity of the NPP / prognostic data from Numerical Weather Prediction (NWP) models of National Weather Services
Example of RODOS calculations during nuclear emergency trainings on Zaporizzhe NPP 22.08.2002
a) Integral concentration of I-131 in air, calculated by RODOS with the use of NWP data
b) calculated by RODOS with the use of single meteorological observation in the point of release
c) Wind streamlines in domain of RODOS’s calculations, calculated by the NWP model MM5, operated in IPMMS NASU
a) b)
c)
Time of release:11-00 UTCTime of NWPAnalyses: 6-00
Measurements: past and current local conditions
NWP data: wide range in space and future time, where no measurements exist
Simultaneous
use by MPPConsistencyMethodology for reconciliation
The problem
Objective
The introduction of data assimilation (DA) techniques in the MPP of the RODOS, acting as diagnostic meteorological model to reconcile the NWP data with the local meteorological stations observations
Choosing method of solution
Strong need in real-time applicability
MPP acts as diagnostic wind model
Applicability to domains with complex geometry
Method of solution:Multivariate optimal interpolation
combined with various meteorological parameterizationsof atmospheric boundary layer (ABL)
and with variational divergence minimizing procedure
Only three dimensional data assimilation (3DDA): Statistical or variational ?
Statistical preferable Variational preferable
1. Calculating first guess fieldCalculated from the NWP data by 1/r2 interpolation:1) 3D fields of velocity, pressure, temperature, humidity2) 2D fields of precipitation, mixing layer height, sensible heat flux, cloud cover,
net radiation (if available from the NWP data)
2. Pre-processing of observations
1) calculation of the net radiation/cloud cover and sensible heat flux in the points of observations from measured values of surface temperature and cloud cover/netradiation (S. Hanna, J. Chang, 1993, van Ulden, Holstag, 1985)2) calculation of the friction velocity and Monin Obukhov length from the measured values of wind velocity and values of sensible heat flux (iterative procedure)3) vertical extrapolation of the measurements of the wind velocity to the vertical levels of MPP
up to the lower 200 m. of the atmosphere (Monin-Obukhov theory, van Ulden, Holstag, 1985)
Cycle of data assimilation
3. Data assimilation1) assimilation of the measured values of cloud cover/net radiation, surfece temperature,
precipitation2) assimilation of the measured (and vertically extrapolated) values of the wind velocities
4. Post-processing1) applying variational divergence minimizing procedure (Sasaki, 1952, Bartzis, et.al., 1998)2) calculation of all other variables needed for ADM using standard meteorological
parameterizations
General optimal interpolation algorithm (Daley, 1991)
Background field: ri
rk - Observations
KkIi 1,1
"a" - analyzed (improved forecast) field; "b" - background (unimproved forecast) field;"T" - true field, "o" - observations
I
j jbf
kjkbf
1)()( rr (1) forward interpolation operator
K
k
I
j jbf
kjkofikW
ibf
iaf 1 1)()()()( rrrr (2) – form of correction, Wik - unknown matrix
fboTiibia εεΩεWrr )()( (3),
)()()(iT
fia
fa
rrri
)()()(kT
fko
fo
rrri
)()()(iT
fib
fb
rrri
I
j jTf
kjkTf
kf 1)()( rrr
assumptions:( ) ( ) 0f f f
T T Tr ro b b b o i ib b ε ε ε ε ε ε ε ε0,fo b ε ε ε (4)
Squaring (3), taking expected values and minimizing with respect to Wi gives: W
T
Ο ΩBΩ F ΩBii (5)TooOTff FTBBB
2B
B
Observation error covariance matrix (CV)
Forward interpolation error CV
Background field error CV
Vector of background field RMS errors
Procedure (1)-(5) is equivalent to minimizing functional:
1
1
T
a o a o
T
a b a b
J
Ωf f O F Ωf f
f f B f f
Assumed statistical structure of the background and measurement errors
Errors of the background field: isotropy, constant rms of each variable
Scalar field:2 2 2 2( ) ( ) exp( )0m r r r Rb ik b b ik b ik (6)
µ - correlation function,R0 – radius of influenceσB – root mean square error
Isotropic vector field,Batchelor, 1953:
Each isotropic homogeneous vector field can be represented as sum of the isotropic homogeneous potential and non-divergent non-correlating vector fields (Obukhov, 1954),
, 2
( ) ( ) ( ) ( )( , ) ( ) ( )
q i q j k i k j
vq vk i j tt qk tt ll
x r x r x r x rr r r r r
r
Let ψ – correspondent stream function, χ – correspondent potential with isotropic distributions:2( ) ( ) ( )ik ik ikm r r r 2( ) ( ) ( )ik ik ikm r r r
2 22
2
/
u
R
ν- ratio of divergent kinetic energy to the total horizontal kinetic energy, R - radius of influence in (r), then (Daley, 1991):
22 2 2 2
2
1( ) (1 )ll
d dr R R
r dr dr
22 2 2 2
2
1( ) (1 )tt
d dr R R
dr r dr
In current work =0
(7)
For all RMS errors of the background field B assumed: B= B(z), assumed also Bu=Bv (9)
(8)
Observation error covariance matrixTooO is assumed to be diagonal with RMS error: O= O(rk); assumed also: Ou=Ov (10)
exp /r R
Multivariate optimal interpolation algorithm for assimilation of wind velocities
Derived using standard OI algorithm (1)-(5) and assumptions (8)-(10)
u ( ) u ( ) uu O B uv O BA i B ir r W W u u v v
v ( ) v ( )r r W Wvu O B vv O Bi iBA u u v v(11)(1)
)( irBuvuvWB
uvuuWIBuu
)( irBuvuvWIB
vvuuWBuv
)( irBuvvvWB
uvvuWIBuu
)(2ir
BvvvvWO
BvvvuWB
uv
(12)(5)
Kkrrr
Klkrr
kOBkiBuuki
Buu
kOBlkBuukl
Buu
1,/),()(
,1,/),(22
22
Note, that in (12) included are only relative errors:
being the key parameter tuning between the observations and background field
2 2B O
2
0 20
( )( , ) exp( / ) 1 1 i j
uu i j
x xrr r r R
R r
2
0 20
( )( , ) exp( / ) 1 1 i j
vv i j
y yrr r r R
R r
0 20
( )( )( , ) exp( / ) i j i j
uv i j
x x y yrr r r R
R r
Determination of
Link can be established with the approach for weighting coefficient used in the MPP “CALMET” of CALPUFF system (Scire, et. al., 1999)
In CALMET: (1 )A O B O OW W f f f
From statistics for one-point measurements:
(12)
2
21
f foB O bfA
B O
HORI
HFINE
HCOARSE
Terrain height
ORICOARSECOARSEHHRMSRMS
FINECOARSEFINE
HHRMSRMS
SW
ZW
OW
20.1,2/
iz
finRMSMIN
ziW n
RMSori
RMSfin
RMSS
W
0
2 20 0/(1 )B O W W
Domain of calculations for ETEX experiment (300x300 km.)
Statistical characteristics of wind field improvement
-1.35.918.76.74.269.55biasd, dec. deg.
424856313131rmsd, dec. deg.
-0.37-0.180.5-0.20.251.06biasu, m/s
2.01.942.212.532.533.19rmsu, m/s
IOSETEX2
OIETEX 2
ECMWF prognose
ETEX 2
IOSETEX 1.
OIETEX 1.
ECMWF analyses, ETEX1
Variable
-1.35.918.76.74.269.55biasd, dec. deg.
424856313131rmsd, dec. deg.
-0.37-0.180.5-0.20.251.06biasu, m/s
2.01.942.212.532.533.19rmsu, m/s
IOSETEX2
OIETEX 2
ECMWF prognose
ETEX 2
IOSETEX 1.
OIETEX 1.
ECMWF analyses, ETEX1
Variable
1.72, -9.22.63, -8.4MM5, dx = 4 km
2.23, -2.58, -RAMS, dx = 0.33 km
0.93, -3.63, -RAMS, dx = 1.32 km
With data assimilationrmsu (m/s), biasd (dec. degree)
Without data assimilationrmsu (m/s), biasd (dec. degree)
Model, scale of grid
1.72, -9.22.63, -8.4MM5, dx = 4 km
2.23, -2.58, -RAMS, dx = 0.33 km
0.93, -3.63, -RAMS, dx = 1.32 km
With data assimilationrmsu (m/s), biasd (dec. degree)
Without data assimilationrmsu (m/s), biasd (dec. degree)
Model, scale of grid
For comparison effect of 4DDA in some models (Seaman, 2000)
Vertical wind profilesa1) 24 October, 12 h
0
100
200
300
400
500
600
0 5 10 15 20
U, m/s
a2) 24 October, 12h
0
100
200
300
400
500
600
150 200 250 300
D, dec. degree
b1) 24 October 18h
0
100
200
300
400
500
600
0 5 10 15 20
U, m/s
b2) 24 October 18h
0
100
200
300
400
500
600
150 200 250 300
D, dec. degree
Vertical profiles of the wind velocity a1)-b1) and of the wind direction a2)-b2), calculated by the MPP with the use of observations (■), with the use of the ECMWF data only (▲), measured by the sodar (,line)Sodar measurements were notused in data assimilationa1), a2) – 12-00 UTC 24/10/1994.b1), b2) – 18-00 UTC24/10/1994.
a1)
a2)
b1)
b2)
Vertical wind profilesc1) 25 October, 00h
0
100
200
300
400
500
600
0 5 10 15 20
U, m/s
d1) 25 October, 06h
0
100
200
300
400
500
600
0 5 10 15 20
U, m/s
z, m
c2) 25 October, 00h
0
100
200
300
400
500
600
150 200 250 300
D, dec. degree
d2) 25 October, 06h
0
100
200
300
400
500
600
150 200 250 300
D, dec. degree
z, m
Vertical profiles of the wind velocity a1)-b1) and of the wind direction a2)-b2), calculated by the MPP with the use of observations (■), with the use of the ECMWF data only (▲), measured by the sodar (,line)Sodar measurements were notused in data assimilationa1), a2) – 00 UTC 25/10/1994.b1), b2) – 06-00 UTC25/10/1994.
Comparison of friction velocity and kinematic heat flux
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
23.5 24 24.5 25 25.5 26 26.5
time, days-0.1
-0.05
0
0.05
0.1
0.15
0.2
0.25
0.3
23.5 24 24.5 25 25.5 26 26.5
time, days
U*, m/s
<w
'T '>
, mK
/s
a)T
ime
depe
nden
ce o
f th
e fr
ictio
n ve
loci
ty .
b)T
ime
depe
nden
ce o
f th
e ki
nem
atic
hea
t flu
xD
ots
- m
easu
red
valu
es (
soni
c an
emom
eter
at t
he M
onte
rfil)
, sol
id b
lack
line
-
calc
ulat
ed d
ata
with
the
use
of D
A p
roce
dure
s, d
ashe
d lin
e -
calc
ulat
ed w
ith th
e us
e of
the
EC
MW
F d
ata
only
Mea
sure
men
ts o
f so
nic
anem
omet
er w
ere
not u
sed
in D
A p
roce
dure
0 100000
x , m
0
100000
y, m
0 100000
x , m
0
100000
y, m
0 100000
x , m
0
100000
y, m
0 100000
x , m
0
100000
y, m
a) b)
d) c)
Ground level wind fields
a) Backgroundfrom ECMWFb) IOSc) IOd) Measured
2D fields of net radiation and cloud cover
50000 100000 150000 200000 250000 300000 350000
x , m
N et rad ia tion, w ith D A , 25 O ctober 00h.
50000
100000
150000
200000
250000
300000
350000
y, m
-75
-70
-65
-60
-55
-50
-45
-40
-35
-30
50000 100000 150000 200000 250000 300000 350000
x , m
N et rad ia tion, w ithout D A , 25 O ctober 00h.
50000
100000
150000
200000
250000
300000
350000
y, m
-80
-75
-70
-65
-60
-55
-50
-45
-40
-35
-30
50000 100000 150000 200000 250000 300000 350000
x , m
C loud cover (oktas), w ith D A , 25 O ctober 00h.
50000
100000
150000
200000
250000
300000
350000
y, m
50000 100000 150000 200000 250000 300000 350000
x , m
C loud cover (oktas),w ithout D A , 25 O ctober 00h.
50000
100000
150000
200000
250000
300000
350000
y, m
0
1
2
3
4
5
6
7
8
0
1
2
3
4
5
6
7
8
Further developments
1. D
A d
eve
lop
ed n
eed
mo
re e
nha
nce
d c
apa
bili
ty to
dea
l with
flow
s in
com
ple
x g
eom
etri
esN
ow w
e re
ly o
n: 1
) qu
ality
of
the
NW
P m
odel
; 2)
rela
tions
for
B
2 / O
2 ; 3
)div
erge
nce
min
imiz
ing
proc
edur
e
Wha
t fur
ther
can
be
done
?
1)A
dvan
ced
met
eoro
logi
cal
para
met
eriz
atio
ns f
or p
re-p
roce
ssin
g of
obs
erva
tions
in
com
plex
geo
met
ries
: i.e
., fo
r ca
lcul
atio
n of
fl
ux p
aram
eter
s (s
ensi
ble
heat
flu
x an
d ot
her,
Bar
low
, B
elch
er,
et.
al.,
2000
), f
or e
stim
atin
g m
ixin
g he
ight
and
ver
tical
ex
trap
olat
ion
of
win
d/te
mpe
ratu
re
mea
sure
men
ts
in
com
plex
ge
omet
ries
(e
.g.,
Zili
tinke
vich
, 200
4)
2)R
evis
ing
corr
elat
ion
func
tions
(6)
, (8)
to a
ccou
nt f
or a
niso
trop
y in
trod
uced
by
com
plex
geo
met
ries
2.1)
sim
ples
t app
roac
h (u
sed
in D
A o
f th
e so
me
mes
osca
le m
odel
s, e
.g. M
M5,
S
eam
an, 1
998)
is to
use
for
m: µ
(ri ,
rj )
=µ
1(|r
j-ri |
)µ2(z
)µ3(
z b)2.
2) e
nsem
ble
met
hod
(Zup
ansk
i, an
d ot
her)
2.3)
may
be
som
ethi
ng w
ill b
e kn
own
from
nat
ure
?
3) M
inim
izin
g ab
ovem
entio
ned
cost
fun
ctio
nal w
ith c
onst
rain
ts: v
aria
tiona
l app
roac
h (P
enen
ko a
nd o
ther
)
Variational approach for 3DDA
Min
imiz
e fu
nctio
nal:
(1)
with
con
stra
ints
(2)1 2( ( ), ( ),..., ( ))nJ dF f f f
r r r
1,..., 0, 1,i nf f i l
For
inst
ance
, “di
verg
ence
min
imiz
ing”
: 1
u v w
x y z
min
imiz
ing
Lag
rang
ian:
B=
0;
;
22 2 22 0 0 01 2 12 2 2
2
u v w
x y zx y z
G
ener
ally
, ver
y fe
w c
ases
, whe
n L
agra
ngia
n si
mpl
ifie
s si
tuat
ion,
one
mor
e is
:
222 2 2 2
( , , , ) 1 0 1 0 2 ( , , )L u v w u u v v w w dVV
u v wx y z
x y z
2 222 2 2( , , ) 1 0 1 0 2 0J u v w u u v v w w dV
V
adju
stm
ent o
f th
e w
ind
velo
citie
s pe
rtur
batio
ns in
the
oute
r re
gion
of
the
cano
py f
low
:
1u pU
x x
1w p
Ux z
0u w
x z
whe
n z>
>l
22( , , ) 0 0J u v w u u w w dV
V
01u w
x z
02
u w
z x
0 02 21 1
1 1; ;
2 2u u v v
x y
2 20 01 1
2 22
u w
x z x z
2 2
0 02 22 2
2u w
x z z z
2
0
2
0
1
2
,1 2( , , )
,
,
L u wV
u u
w w
u wx z
x x
u wx z dV
z x
Variational approach for 3DDA
1 1T T
a o a o a b a bJ Ωf f O F Ωf f f f B f f
Gen
eral
cas
e fo
r 3D
DA
: min
imiz
e fu
nctio
nal (
the
sam
e as
in O
I):
(3)
with
con
stra
ints
:
0, 1,ai i n f (4)
Pro
blem
(3)
-(4)
usu
ally
can
be
solv
ed n
umer
ical
ly u
sing
sta
ndar
d ap
proa
ches
(e
.g.,
pena
lty +
des
cent
alg
orith
ms
or o
ther
mor
e ad
vanc
ed).
T
he m
ain
com
plex
ity o
f th
e pr
oble
m is
cau
sed
by th
e ch
oice
of
the
con
stra
ints
(4)
Conclusions
1.M
etho
dolo
gies
for
the
ass
imila
tion
of t
he o
bser
vatio
ns o
f w
ind
velo
citie
s an
d ot
her
in th
e M
PP
of
the
ER
S s
yste
m R
OD
OS
wer
e de
velo
ped
2.T
he m
ultiv
aria
te o
ptim
al i
nter
pola
tion
sche
me
com
bine
d w
ith t
he r
elat
ions
for
th
e w
eigh
ting
coef
fici
ent
used
in
M
PP
C
AL
ME
T
was
fo
r th
e fi
rst
time
impl
emen
ted
as a
3D
DA
sch
eme
in th
e M
PP
of
the
real
-tim
e E
RS
sys
tem
3.C
ompa
riso
ns
of
the
mod
el
resu
lts
with
th
e m
eteo
rolo
gica
l m
easu
rem
ents
pe
rfor
med
in
the
ET
EX
exp
erim
ents
sho
wed
goo
d ag
reem
ent
of c
alcu
late
d va
lues
with
mea
sure
men
ts a
nd i
mpr
ovem
ent
of t
he f
irst
gue
ss f
ield
pro
duce
d us
ing
the
NW
P r
esul
ts w
ith th
e us
e of
the
3DD
A p
roce
dure
s4.
Fur
ther
dev
elop
men
t of
the
dat
a as
sim
ilatio
n pr
oced
ures
for
the
MP
Ps
of t
he
ER
S
shou
ld
be
perf
orm
ed
for
prod
ucin
g m
ore
phys
ical
ly
cons
iste
nt
met
eoro
logi
cal f
ield
s w
hen
appl
ied
in c
ompl
ex g
eom
etri
es
Acknowledgements
The
pre
sent
wor
k ha
s be
en f
ully
sup
port
ed b
y th
e E
urop
ean
Com
mis
sion
thr
ough
the
EU
RA
TO
M g
rant
in
conn
ectio
n to
the
E
urop
ean
Pro
ject
"R
OD
OS
Mig
ratio
n".