Transcript Beekmann-MonteCarlo

Part 1
Monte Carlo uncertainty evaluation
of emission reduction scenarios
constrained by observations from the
ESQUIF campaign
M. Beekmann (LISA),
C. Derognat (Aria-Technologies)
Part 2
Extension of CHIMERE to Eastern
Europe and evaluation with surface
and satellite data
I. Konovalov (Institute of Appplied Physics, Nizhny
Novgorod) M. Beekmann (LISA)
R. Vautard (LMD/IPSL)
A. Richter (IUP, University of Bremen)
J. Burrows (IUP, University of Bremen)
,
What is the uncertainty in the simulation
of emission reduction scenarios ?
 Case of Paris agglomeration
Monte Carlo uncertainty analysis

Model output uncertainty due to uncertainty in input
parameters

Constraint by measurements (ESQUIF campaign)
(Bayesian Monte Carlo uncertainty analysis)

Reduced uncertainty
METHODOLOGY (1)
SET-up of the CHIMERE model
for the Paris region (version 2002)
 Domain 150 km x 150 km with 6
km horizontal resolution
 5 vertical levels from surface to
~3 km
 Forced by ECMWF first guess or
forecast
 Gas phase chemistry:
MELCHIOR with 82
compounds, 338 reactions
 Emissions, refined for regional
scale from AIRPARIF, also
biogenic
 Boundary conditions: from
CHIMERE at continental scale
OX, NOy 16/7/99 14h POI6
METHODOLOGY (2)
Definition of the probability density function for input
parameters
EMISSIONS :
anthropogenic VOC
anthropogenic NOx
biogenic VOC
 40 % (log.,1)
 40 % (log.,1)
 50 % (log.,1)
RATE CONSTANTS
NO + O3
NO2 + OH
NO + HO2
NO + RO2
HO2 + HO2
RO2 + HO2
RH + OH
CH3COO2 + NO
CH3COO2 + NO2
PAN + M










10
10
10
30
10
30
10
20
20
30
%
%
%
%
%
%
%
%
%
%
Hanna et al, 1998
as for VOC
Hanna et al, 1998, 2001
:
(log.,1)
(log.,1)
(log.,1)
(log.,1)
(log.,1)
(log.,1)
(log.,1)
(log.,1)
(log.,1)
(log.,1)
Atkinson
Atkinson
Atkinson
Atkinson
Atkinson
Atkinson
Atkinson
Atkinson
Atkinson
Atkinson
et
et
et
et
et
et
et
et
et
et
al,
al,
al,
al,
al,
al,
al,
al,
al,
al,
1997
1997
1997
1997
1997
1997
1997
1997
1997
1997
PHOTOLYSIS FREQUENCIES + RADIATION :
actinic flux
 10 % (log.,1)
see text
J(O3 -> -> 2 OH)
 30 % (log.,1)
DeMore et al, 1997
J(NO2->NO+O3)
 20 % (log.,1)
DeMore et al, 1997
J(CH2O->CO+2 HO2)  40 % (log.,1)
DeMore et al, 1997
J(CH3COCO-> ....)
+ 50 % (one sided, 1)
S 95, RM 96
J(carbonyl compound from o-xylene) 40 % (log.,1 Atkinson al, 1997
METEOROLOGICAL PARAMETERS:
zonal wind speed
 1 m/s (absolute,1)
meridional wind speed  1 m/s (absolute,1)
mixing layer height
 20 %
(log.,1)
temperature
 1.5 K (absolute,1)
relative humidity
 20 %
(log.,1)
vertical mixing coefficient  50 % (log.,1)
deposition velocity
 25 %
(log.,1)
see text
see text
see text
Hanna et al, 1998
after Hanna et al, 1998/2001
see text
Hanna et al, 1998/2001
METHODOLOGY (3)
Constraints from ESQUIF observations
From circular flights (DIMONA, MERLIN)
 DOX, DNOy, DNOx, (DVOC)
DC = C (plume) – C (background)
From airquality network (AIRPARIF)
 DOX = OX (urban) – OX (background)
Flight tracks around the Paris agglomeration during ESQUIF
METHODOLOGY (3)
Constraints from ESQUIF observations
From circular flights (DIMONA, MERLIN)
 DOX, DNOy, DNOx , (DVOC)
DC = C (plume) – C (background)
METHODOLOGY (4)
mathematical formulation of the constraint
For each Monte Carlo simulation k:
Likelihood L for model output Yk to be correct for observations Oi
(Bayesian Monte Carlo analysis Bergin and Milford, 2000):
1
(Oi – Yk,i) 2
L(YkY | Oi) = _____________ EXP [ -0.5
(2p0.5 i
_______________
 i2
]
L(Yk | O) = L(Yk,,1 | O1) * L(Yk,2 | O2) * …….
Measurement errors i of observations Oi are assumed as
 normally distributed
 independent
They stem from
 instrumental errors
 uncertainty in representativity for model grid
METHODOLOGY (5)
Simulations performed
 For 3 days in POI’s 2 and 6: August7, 1998 and July 16,17
 500 Monte Carlo simulations with base line emissions
 500 Monte Carlo simulations with reduced emissions



- 50 % anthropogenic VOC
- 50 % anthropogenic. NOx
- 50 % anthro. VOC + NOx
RESULTS (1)
• Cumulative
probability plots
Surface O3 maxima for
baseline and 50% reduced
emissions
With (____) and without
(- - - -) constraint
RESULTS (2)
Surface O3 maxima for baseline and 50% reduced emissions
Chemical regime averaged
over the pollution
plume:
Difference in surface O3
between a
 NOx emissions –50 %
and a
 VOC emissions –50%
scenario
Positive values : VOC limited
chemical regime
Average over 1998/1999 :
VOC sensitive or intermediate
chemical regime
(thesis C. Derognat)
RESULTS (3)
RESULTS (4)
OH averaged over the
pollution plume
at 14 UT (layer 2 50-600
m):
RESULTS (5)
A posteriori
and a priori
probability of
input parameters :
NOx and VOC
emissions
CONCLUSIONS
 Uncertainty in simulated max. ozone
(for baseline and reduced emissions) reduced by a factor 1.5 to 3
due to measurement constraint
 Uncertainty in VOC limited regime is reduced for two days,
shift from slightly VOC limited to slightly NOx limited for
anaother day
 For OH, the uncertainty is less reduced, but very low values are
rejected, remaining uncertainty factor 1.5 – 2.5
 Weighting procedure through likelihood function changes
distribution in input parameters namely NOx emissions
Limitations of this study:
 Uncertainty in model formulation is neglected (transport, model chemistry)
 Uncertainty in the definition of pdf’s for input parameters
 Uncertainty in error distribution of observations
(covariance always zero ?)
Perspectives :
 Application to continental scale
 Application to air quality forecast
Part 2
Extension of CHIMERE to Eastern
Europe and evaluation with surface and
satellite data
I. Konovalov (Institute of Appplied Physics, Nizhny Novgorod)
M. Beekmann (LISA)
R. Vautard (LMD/IPSL)
A. Richter (IUP, University of Bremen)
J. Burrows (IUP, University of Bremen)
,
Model set up
 Domain covering EU to Ural +
Mediterranean regions with 0.5 °
horizontal resolution
 8 vertical levels from surface to 500
hPa
 Forced by NCEP forecast (2.5°) and
MM5 (1° res.)
 Gas phase chemistry: MELCHIOR
reduced
 Emissions from EMEP and EDGAR,
if needed
 Boundary conditions: from
MOZART
Time series
Error statistics
Comparison between GOME
and CHIMERE derived
tropospheric NO2 columns,
June – August 1997
University of Bremen,
GOME version V2
320 * 40 km resolution
I. B. Konovalov, M. Beekmann,
R. Vautard, J. P. Burrows,
A. Richter, H. Nüß,
N. Elansky, ACP, 2005
CHIMERE tropospheric NO2 columns
versus
GOME tropospheric NO2 columns
Average June – August 1997
Western Europe
Eastern Europe
Slope = 0.75
R = 0.91
Slope = 0.70
R = 0.77
differences in GOME / CHIMERE tropospheric NO2 columns
versus
tropospheric NO2 columns (1015mol.)
Western Europe
 Random error in monthly mean (in a spatial sens) is mainly of multiplicative
nature (25-30%), no attribution to GOME or CHIMERE possible
differences in GOME / CHIMERE tropospheric NO2 columns
versus
tropospheric NO2 columns (1015mol.)
Eastern Europe
 Random error in monthly mean (in a spatial sens) is less clearly of multiplicative
nature for Eastern Europe than for Western Europe
CONCLUSIONS
 CHIMERE domain has been extended to Eastern EU and
Mediteranean region
 Correlation with surface O3 obs. larger in WE (>80%)
than in Central and EE <60-70%)
 Comparison with GOME tropospheric NO2 :
* No bias
* slope 0.70-0.75
* multiplicative spatial random error 15% EE – 30% WE