chemical modelling & data assimilation
DESCRIPTION
Chemical Modelling & Data Assimilation D. Fonteyn, S. Bonjean, S. Chabrillat, F. Daerden and Q. Errera Belgisch Instituut voor Ruimte – Aëronomie (Belgian Institute for Space Aeronomy) BIRA - IASB. >> OUTLINE. Introduction What is chemical data assimilation? - PowerPoint PPT PresentationTRANSCRIPT
A NICE PLACE 1
Chemical Modelling & Data Assimilation
D. Fonteyn, S. Bonjean, S. Chabrillat, F. Daerden and Q. Errera
Belgisch Instituut voor Ruimte – Aëronomie
(Belgian Institute for Space Aeronomy)
BIRA - IASB
A NICE PLACE 2
>> OUTLINE
• Introduction• What is chemical data assimilation? • Why do we need chemical data assimilation?• 4D –VAR chemical data assimilation system• Physical consistency, Self consistency, Independent
observations• Added value• Inverse modelling: emission estimations
A NICE PLACE 3
>> OUTLINE
Belgian Assimilation System of Chemical Observations
from Envisat (BASCOE) http://bascoe.oma.be
IMAGES
A NICE PLACE 4
>> Introduction
• Focus on the Stratosphere:– Chemical processes are well understood: high level of
confidence in modelling results. (?)
– Mature remote sensing technology (UARS, ENVISAT, SAGE, CRISTA, POAM …)
• If models are perfect, no data assimilation is needed
A NICE PLACE 5
>> Introduction >> Overview chemistry
Gas phase chemistryChapman Cycle
O2 + h 2O
O2 + O + M O3 + M
O3 + O 2O2
O3 + h O2 + O
Catalytic cyclesHydrogen radicals (HOx)
OH + O H + O2
H + O3 OH + O2
Net: O + O3 2 O2
OH + O H + O2
H + O2 + M HO2+ M
HO2 + O OH + O2
Net: O + O3 O2
HO2 + O OH + O2
OH + O3 HO2 + O2
Net: O + O3 2 O2
Hydrogen Source Gases: H2O, CH4
• Long term trends• HOx chemistry in the upper
stratosphere and mesosphere
A NICE PLACE 6
>> Introduction >> Overview chemistry
2. Nitrogen radicals (member of NOy)NO2 + O NO +
O2
NO + O3 NO2 + O2
Net: O + O3 2 O2
3. Chlorine radicals (member of Cly) ClO + O Cl +
O2
Cl + O3 ClO + O2
Net: O + O3 2 O2
Chlorine Source Gases: Organic Chlorine• Long term trends• Cly partitioning (in the lower
stratosphere: aerosols )
Nitrogen Source Gas: N2O (and …)• Long term trends• NOy partitioning (in the lower
stratosphere: aerosols )
A NICE PLACE 7
>> Chemical data assimilation
Chemical data assimilation
• Inert tracer assimilation
• Tracer with parameterized chemistry assimilation
• Multiple species with chemical interactions
Necessity
A NICE PLACE 8
>> Why chemical data assimilation >> Model shortcomings
TOMS total ozone 28 August 2003
A NICE PLACE 9
>> Why chemical data assimilation >> Model shortcomings
Free model total ozone 28 August 2003, 12 UTC
A NICE PLACE 10
>> Why chemical data assimilation >> Model shortcomings
-90 -60 -30 0 30 60 90
10-1
100
101
102
103
Latitude
Pres
sure
[hPa
]
0.2
0.4
0.6
0.8
1
1.2
1.4
1.6
HALOE CH4 monthly gridded zonal mean, August 2003
ppmv
A NICE PLACE 11
>> Why chemical data assimilation >> Model shortcomings
-90 -60 -30 0 30 60 90
10-1
100
101
102
103
Latitude
Pres
sure
[hPa
]
0.2
0.4
0.6
0.8
1
1.2
1.4
1.6
HALOE CH4 monthly gridded zonal mean, August 2003Free Model Run co-located, isolines
ppmv
A NICE PLACE 12
-80 -60 -40 -20 0 20 40 60 800
1
2
3
4
5
6
Latitude
Mea
n A
ge o
f Air
(yea
r)ERA40/ANAERA40/FOROPER/ANAOPER/FORCO2 observationsSF6 observations
>> Why chemical data assimilation >> Model shortcomings
Problem: input dynamics, confirmed by mean age of air experiment
A NICE PLACE 13
>> Why chemical data assimilation >> Model shortcomings
CH4
GEM STRATO (MSC) with BASCOE chemistry vs. BASCOE driven by ECMWF• 3 month free model run• Same initial conditions• Matching resolution• Identical chemistry• No Feedback
A NICE PLACE 14
CH4
BASCOE driven by GEM-STRATO vs BASCOE driven by ECMWF
>> Why chemical data assimilation >> Model shortcomings
A NICE PLACE 15
Ozone
BASCOE driven by GEM-STRATO vs BASCOE driven by ECMWF
>> Why chemical data assimilation >> Model shortcomings
A NICE PLACE 16
Total ozone
BASCOE driven by GEM-STRATO vs BASCOE driven by ECMWF
>> Why chemical data assimilation >> Model shortcomings
A NICE PLACE 17
Model Shortcomings:• Effect of dynamical assimilation• Effect of different dynamical assimilation systems
• Dynamics driven shortcomings
• Chemical modelling shortcomings (not shown)
>> Why chemical data assimilation >> Model shortcomings
A NICE PLACE 18
>> 4D – VAR
N
iii
oi
Tii
obTb tHttHtttttJ1
100
1000 )](x[)(yR)](x[)(y
21)(x)(xB)(x)(x
21
4D-var assimilation : find x(t0) minimizing J
With the constraint
x(t0): control variable n 5.6 106
xb: a priori state of the atmosphere ( background)yo(ti): observations, de dimension p 5 104 (-7 105)x(ti): model state H: observational operatorM: model operatorB: background error covariance matrixR: observational error covariance matrix
)]([)( tMdt
td xx
A NICE PLACE 19
>> 4D – VAR >> BASCOE
• Model (3D - Chemical Transport Model)– horizontal: 3°.75 x 3°.75 (96 x 49 pts); vertical: 37 pressure levels, surface → 0.1 hPa
(subset of ECMWF hybrid levels, keeping stratospheric levs)– 57 chemical species (control variables), 200 reactions– 4 types of PSC particles (36 size bins): NOT assimilated– Eulerian, driven by ECMWF 6h analyses/forecast– advection by Lin & Rood (1996) with 30’ time step
• Assimilation set-up– Adjoint of chemistry and transport– Assimilation time window: 24 hours– B diagonal; 20 % of first guess distribution (= univariate)– Quality check: 1st climatoligical behaviour; 2 nd first guess based QC
• Observations– ESA Envisat MIPAS L2 products, Near Real Time (NRT) and Offline (OFL)– O3, H2O, N2O, CH4, HNO3, NO2
– Representativeness error: 8.5 %
A NICE PLACE 20
4D – VAR >> BASCOE >> OFL number of observations
Jan03 Mar03 May03 Jul03 Sep03 Nov03 Jan040
1
2
3
4
5
6
7
8
9
10 x 104
Num
ber o
f Obs
.Total validPre-rejectedOIQC-rejected
A NICE PLACE 21
4D – VAR >> BASCOE >> Multi-variate nature
MJD 1315, 2.66 hPa local noon: Relative change w.r.t ozone
0.001 0.01 0.1 1 10
O3
HOX
NOYNO
NO2NO3
N2O5HNO3
CLYCLO
HOCLCLONO2
HCL
BRYBRO
HOBRBRONO2
HBR
N2OCH4H2O
CFC11
0.001 0.01 0.1 1 10
MJD 1315, 29 hPa, local midnight: Relative change w.r.t ozone
0.001 0.01 0.1 1 10
O3
HOX
NOYNO
NO2NO3
N2O5HNO3
CLYCLO
HOCLCLONO2
HCL
BRYBRO
HOBRBRONO2
HBR
N2OCH4H2O
CFC11
0.001 0.01 0.1 1 10
Multi variate nature– Diagonal B– (xa(t0)-xb(t0))– Local noon and local midnight– August, 7, 2003– Full: observed species– Striped: unobserved species
A NICE PLACE 22
4D – VAR >> BASCOE >> Physical consistency
O3 / ANA O3 / FMR
CH4 / ANA CH4 / FMR
N2O / ANA N2O / FMR
O3 / MIPAS / OFL
CH4 / MIPAS / OFL
N2O / MIPAS / OFL
August 5, 2003
35.8 hPa & obs within 1 km
A NICE PLACE 23
4D – VAR >> BASCOE >> Physical consistency
0 50 100 150 200 250 300 350 4000
1
2
CH4
N2O
0 50 100 150 200 250 300 350 4000
1
2
CH4
N2O
0 50 100 150 200 250 300 350 4000
1
2
CH4
N2O
Tracer correlations:CH4 vs N2O (Aug 5)TropicalSouth polar
MIPAS DATA
Co-located FMR
Co-located analysis= correlationNeeds validation
A NICE PLACE 24
4D – VAR >> BASCOE >> Self – consistency
-5 0 50
0.5
1O3
-5 0 50
0.5
1H2O
-5 0 50
0.5
1CH4
-5 0 50
0.5
1N2O
-5 0 50
0.5
1HNO 3
-5 0 50
0.5
1NO2
• OmF:– Observation – first guess– Normalized by R
– = Gaussian distribution– OFL– NRT– FMR– OFL vs NRT– Consistency – Added value w.r.p FMR
A NICE PLACE 25
4D – VAR >> BASCOE >> Self – consistency >> model improvement
0 2 4 6 8 10 12
10-1
100
101
102
103
Pres
sure
[hPa
]
Mixing ratio [ppmv]
MIPAS NRTNRT analysisNRT FMROFL FMR
NRT results:Ozone @ 1 hPa underestimated
– Analysis = free model– Model not constrained– O2 main source of O3
– O2 not a control variable
– JO2 increased by 25 %
– New free model
– Better agreement
A NICE PLACE 26
4D – VAR >> BASCOE >> Self – consistency
Self – consistency 4D – VAR:E[Janalysis] = p/2Time series Janalysis/p
NRTOFL
Monitoring capability
A NICE PLACE 27
4D – VAR >> BASCOE >> Self – consistency >> monitoring
Monitoring capabilityDaily mean MIPAS ozone, [-10,10]
at 14 hPa
Janalysis transients correlate with ozone daily mean transients
A NICE PLACE 28
0 0.5 1 1.5 2
x 10-6
10-1
100
101
102
Pre
ssur
e [h
Pa]
NRT CH4 [vmr]
0 0.2 0.4 0.6 0.8 1
x 10-5
10-1
100
101
102
Pre
ssur
e [h
Pa]
NRT H2O [vmr]
0 0.5 1 1.5 2
x 10-6
10-1
100
101
102
OFL CH4 [vmr]
0 0.2 0.4 0.6 0.8 1
x 10-5
10-1
100
101
102
OFL H2O [vmr]
4D – VAR >> BASCOE >> Example
Illustrative example:August 5, 2003Lat: -38.6° Lon: 83.3°
• NRT vs OFL data• Quality check
– Pre-check– OI qc
• First guess• Analysis
• At 1 hPa: methane rich tropical air, and tropical dry air
A NICE PLACE 29
0 0.5 1 1.5 2
10-1
100
101
102
103
VMR [ppmv]
Pres
sure
[hPa
]
20030805 Lat:-40 Lon:122
0 2 4 6 8 10
10-1
100
101
102
103
VMR [ppmv]
Pres
sure
[hPa
]
20030805 Lat:-40 Lon:122
4D – VAR >> BASCOE >> Independent observations
Independent observations:HALOE v19Periode: August 20031. Individual profiles2. Statistics
Individual profile
OFL analysisHALOEFree Model run
CH4
H2O
A NICE PLACE 30
-40 -30 -20 -10 0 10 20 30 40
10-1
100
101
102
103
Relative difference [%]
Pres
sure
[hPa
]
-40 -30 -20 -10 0 10 20 30 40
10-1
100
101
102
103
Relative difference [%]
Pres
sure
[hPa
]
-40 -30 -20 -10 0 10 20 30 40
10-1
100
101
102
103
Relative difference [%]
Pres
sure
[hPa
]
4D – VAR >> BASCOE >> HALOE August 2003
H2O CH4
O3(HALOE-BASCOE)/HALOE
OFL analysisNRT analysisHALOE error
A NICE PLACE 31
0 5 10 15 20
10-1
100
101
102
103
Mean volume mixing ratio [ppbv]
Pres
sure
[hPa
]
0 0.5 1 1.5 2 2.5 3 3.5
10-1
100
101
102
103
Mean volume mixing ratio [ppbv]
Pres
sure
[hPa
]
4D – VAR >> BASCOE >> HALOE August 2003
Mean HALOE and BASCOE co-located profiles:
OFL analysisNRT analysisHALOE
Ozone model bias & 4D – VAR chemical coupling reduces Cly
HCl
NOx
A NICE PLACE 32
-90 -60 -30 0 30 60 90
10-1
100
101
102
103
Latitude
Pres
sure
[hPa
] 0.4
0.81.2
0.2
0.4
0.6
0.8
1
1.2
1.4
1.6
4D – VAR >> BASCOE >> Added value
HALOE and BASCOE co-located gridded zonal monthly mean
OFL analysisFMRHALOE
OFL analyis vs Free Model
(Schoeberl et al., JGR 2003)
A NICE PLACE 33
TOMS total ozone 28 August 2003
4D – VAR >> BASCOE >> Added value
A NICE PLACE 34
Analysis total ozone 28 August 2003, 12 UTC
4D – VAR >> BASCOE >> Added value
A NICE PLACE 35
4D – VAR >> BASCOE >> Added value >> Chemical forecasts
The operational implementation with NRT MIPAS allows to produce chemical forecasts
Examples with verification
A NICE PLACE 36
4D – VAR >> BASCOE >> Added value Operational implementation
A NICE PLACE 37
4D – VAR >> BASCOE >> Added value
A NICE PLACE 38
4D – VAR >> BASCOE >> Conclusions
4D –VAR chemical data assimilation system– Multi-variate nature of 4D – VAR – Benefit– Model bias sensitivity
• Overall Consistency• Independent observations• Added value (non-exhaustive)
– Monitoring– Bias detection– Correction for dispersive dynamics– Chemical forecasts
• Potential related to efforts
A NICE PLACE 39
>> Inverse modelling
Inverse modelling at BIRA – IASB
J. – F. Muller & J. Stavrakou
Belgisch Instituut voor Ruimte – Aëronomie
(Belgian Institute for Space Aeronomy)
BIRA - IASB
A NICE PLACE 40
Focus: Tropospheric reactive gases (ozone precursors CO, NOx,
non-methane VOCs)
>> Inverse modelling
A NICE PLACE 41
>> Inverse modelling
J(f)=½Σi (Hi(f)-yi)T E-1(Hi(f)-yi)+ ½ (f-fB)TB-1(f-fB)
observationsModel operator
acting on the controlvariables
first guess valuefor the controlparameters
Matrix of errorson the emission
parametersMatrix of errors
on the observations
A NICE PLACE 42
• Find best values of emission parameters, i.e. minimize the cost function
• Previous studies for reactive gases (CO, NOx, CH2O) inverted for a small number of emission parameters (big-region approach)
• Most previous studies used a linearized CTM, (i.e. OH unchanged by emission updates) straightforward minimization of the cost (matrix inversion)
• Non-linearity is best handled using the adjoint model technique (Muller & Stavrakou 2005) also used in 4D-Var assimilation
• This technique allows also to perform grid-based inversions
>> Inverse modelling
A NICE PLACE 43
Grid – based inversion• Observations used: CO columns from
MOPITT (05/2000 – 04/2001)• Model used: IMAGES, 5°x5° (Müller and Stavrakou 2005)• Number of control parameters >> number of independent
observations need additional information : correlations between errors on
a priori emissions, estimated based on country boundaries, ecosystem distribution, geographical distance
>> Inverse modelling
A NICE PLACE 44
>> Inverse modelling