theworldweatheropenscienceconference 16&?&21 ... benjamingaubert 1! jérôme!barré...
TRANSCRIPT
Global CO data assimila.on for emissions and trends analysis
Benjamin Gaubert1
Jérôme Barré1, Helen Worden1, David Edwards1, Louisa Emmons1, Simone Tilmes1, Arthur Mizzi1, Avelino Arellano2, Jeffrey Anderson3 and Nancy Collins3
1NCAR, Atmospheric Chemistry Division 2University of Arizona, Tucson USA
3NCAR/IMAGe, InsOtute for MathemaOcs Applied for Geo-‐Sciences
The World Weather Open Science Conference 16 -‐ 21 August 2014
Global models
underes.mate CO observa.ons especially in the Northern Hemisphere
Mo.va.ons
Annual cycle of MOPITT @ 500hPa (red) vs MulOmodel mean (leX) and individual model (right)
Shindell et al. 2006
q But trends are nega.ve over more than 10 years of MOPITT measurement Ø Trends in CO concentra.ons are not really consistent with trends in
inventories Ø Large interannual variability driven by Biomass Burning (BB)
Mo.va.ons
12-‐month running averages for N. Hemisphere total column CO measurements normalized by the 08/2008–07/2009 average CO column for each instrument.
Worden et al. 2013
Using mul.spectral MOPITT retrieval and data assimila.on tools
q Determine the global CO budget Ø CO is a primary pollutant (good tracer of combus.on)
² Anthropogenic ² Biomass Burning ² Biogenic / ocean
Ø CO is also a secondary pollutant involve in the chemical oxida.on cycle of CH4 and NMVOCS
Ø Sink is mainly CO + OH
q Main error sources § Transport § Ini.al condi.ons § Chemistry CH4-‐OH-‐VOCS-‐NOx-‐O3 § Emissions
Constraints
q Data assimila.on: Error minimiza.on § Transport: assimila.ng meteorological data § Ini.al condi.ons: Op.miza.on of [CO] assimila.ng CO data § Chemistry:
Ø Impact through chemistry in the model forecast Ø Op.miza.on of [VOCS], [HOx], [CH4 ]assimila.ng CO data using
ensemble correla.ons § Emissions:
Ø Op.miza.on of CO emissions assimila.ng CO data Ø Op.miza.on of VOCS emissions assimila.ng CO data
q Global CO analysis using CAM-‐CHEM with DART-‐EAKF
Ø To produce 6-‐hourly analysis (both meteorology and CO) Ø within the scope of doing an analysis of the whole decade 2002-‐2014
q Inves.gate the state augmenta.on approach
Ø To evaluate interac.on with meteorology and ozone pollu.on (talk of J. Barré)
Ø To es.mate CO emissions Ø To es.mate contribu.ons from chemistry, adjustment of chemical species
Assimila.on of MOPITT CO profiles in CAM-‐CHEM
www.image.ucar.edu/DAReS/DART
Anderson et al. 2009, Raeder et al. 2012 Arellano et al. 2006, 2010, Barré et al. 2014 in prep
What are the impact of MOPITT CO assimila.on in a fully coupled system ? Mul.spectral system allow for unique impact on the surface, lower layers of the
troposphere
Data assimila.on components : Community Earth System Model (CESM) Community Atmosphere Model for chemistry (CAM-‐CHEM)
(Lamarque et al 2012)
q CESM1_1_1 / CAM5 physics q COMPSET F2000_MOZMAM_CN offline ocean q SpaOal resoluOon : Horizontal : 1.9*2.5⁰ / VerOcal : 30 levels q Emissions:
ü Anthropogenic : MACCity (Granier et al. 2011) ü Biomass Burning / Fire emissions : Daily fire emissions from FINN (Wiedinmyer et al. 2011)
ü Biogenic : offline MEGAN V2.1 (Guenther et al. 2012) no interannual variability
ü CCMI NO2 and BC AircraX emissions q CESM SimulaOon starts in January 1998 § AssimilaOon of meteorological observaOons only
Ø start February 1st 2006 from CESM § AssimilaOon of meteorological observaOons and MOPITT CO
Ø start February 10th 2006
Data assimila.on components: Observa.ons
MOPITT CO mul.spectral retrievals (TIR+NIR)
q MOPITT V5J (Worden et al. 2010, Deeter et al. 2011, 2012, 2013)
Ø CO profiles over 10 levels Ø Improve sensiOvity to to surface / LT with a low bias
q Thinning : Ø DayOme observaOon, between +/-‐ 65⁰, dfs > 0.5
q Super observa.on over the model grid cells (1.9x2.5⁰) Ø Average, ignore observaOon error correlaOon Ø Error variance is reduce according to
Ø To account for representaOveness error CO std (coarse model grid cells)
σ obs =1M
1M
σ ii=1
M
∑
222obsreprMOPITT σσσ +=
March 2006
§ Ensemble : 20 members / i.e 20 CESM simulaOons § Background error :
Ø Space and Ome varying mulOplicaOve inflaOon Ø IniOal T perturbaOon (pert_sd=2)
§ Localiza.on: Gaspari and Cohn 1999 localizaOon funcOon (gaussian shape) Ø Horizontal : half-‐width of 0.2 rad Ø VerOcal : half-‐width of 600 hPa
§ Meteorological data: Ø P, T, U, V, Q : allow cross correlaOons
§ Emissions perturba.ons: (Evensen et al. 2003) Ø CO, Length scale = 2000km, sd=0.4 Ø VOCs, Length scale = 2000km, sd=0.3
Data assimilation components: DART-‐EAKF (Anderson 2001, 2003, 2007, 2009, Raeder et al. 2012)
set-‐up : Meteorological analysis
P T U V Q
P
T
U
V
Q
§ Same setup § Assimilated data : Variable localisa.on
Ø P, T, U, V, Q : allow cross correla.ons Ø MOPITT CO observa.ons affect CO state
§ Localiza.on : Gaspari and Cohn localizaOon funcOon (gaussian shape) Ø Horizontal : half-‐width of 0.1 rad Ø VerOcal : half-‐width of 300 hPa
DART-‐EAKF set-‐up : Meteorological and MOPITT analysis
P T U V Q CO
P
T
U
V
Q
CO
Assimilated observaOons
Mod
el variables
Results : Evalua.on against MOPITT Diagnos.c in observa.ons space
Weighted bias (RCRV mean)
Unbiased RMSE
Spa.al Correla.on
Number of observa.ons
Lower troposphere Middle troposphere Upper troposphere
Met analysis Met & MOPITT prior Met & MOPITT posterior
Ø Huge bias reduc.on
Results : comparison with NASA/Intex B DC8 flight (March 2006)
v All the 6 flights are considered, 4-‐19 March 2006 v Average in in 100hPa bin, outside Mexico city
(Emmons et al. 2010)
Met & MOPITT analysis
Met analysis
Aircraq observa.ons
q Increase of CO mostly in the Northern Hemisphere and in the middle troposphere up to a factor of 2 Ø Consistent with both simula.on error and MOPITT precision
Impact on CO / March 2006
(MOP_MET_analysis – MET analysis) / (MET analysis)
La.tude vs .me Al.tude vs .me
q But no increase or decrease (south-‐east Asia) over biomass burning region
(MOP_MET_analysis – MET_analysis) / (MET analysis) Surface level
Average emissions fluxes Only from biomass burning
Impact on CO (2) / Surface, March 2006
q Increase of CO leads to a decrease of OH and increase of HO2 q Decrease of OH
Ø Decrease of secondary pollutant forma.on : example, formaledehyde (HCHO)
Impact on the chemical fields / Surface, March 2006
HO2 HCHO
Impact on the chemical fields / Surface, March 2006
C3H8 BIGALK
q Increase of CO leads to a decrease of OH and HO2 q Decrease of OH
Ø Increase of VOCS when their sink is VOC + OH
Ø Effec.ve reduc.on of the bias and unbiased errors against MOPITT and Aircraq data of CO
Ø Beser correla.on with MOPITT
Ø Similar corela.ons with Aircraq data although it is much lower scale observa.ons
Ø Modifica.on of the chemical coupling through the CTM simula.on
Global CO analysis using CAM-CHEM with DART-EAKF
q Can use the state augmenta.on to Ø Constraint on VOCs, or other species related to CO
q 3 experiments with ini.al condi.ons op.miza.on
Ø CH4
Ø HOx (OH + HO2)
Ø VOCS
DA Experiments
P T U V Q CO
P
T
U
V
Q
CO
CH4
OH
HO2
VOCS
Assimilated observa.ons
Mod
el variables
DA Experiments : evalua.on against MOPITT
Similar results for CO forecast (and analysis)
Ø Same scores for CO, reduced bias in the boundary layer when correlated with HOx
Results : comparison with NASA/Intex B DC8 flight (March 2006)
v All the 6 flights are considered, 4-‐19 March 2006 v Average in in 100hPa bin, outside Mexico city
(Emmons et al. 2010)
Met & MOPITT analysis (No cor.)
Aircraq observa.ons
Met & MOPITT analysis (cor. CH4)
Met & MOPITT analysis (cor. VOCS)
Met & MOPITT analysis (cor. HOx)
Results : comparison with NASA/Intex B DC8 flight (March 2006)
Correla.on with VOCS reduce the bias against those VOCS HOx and CH4 improve the state variability through chemistry
LOWER BIAS BETTER R # Obs
CO MOP_MET_ANALYSIS_corrHOx MOP_MET_ANALYSIS_corrCH4 935
O3 MOP_MET_ANALYSIS_corrHOx MOP_MET_ANALYSIS 965
CH4 MOP_MET_ANALYSIS_corrHOx MOP_MET_ANALYSIS 905.0
CH2O_NCAR MOP_MET_ANALYSIS_corrVOCS MOP_MET_ANALYSIS_corrCH4 697.0
C2H6 MOP_MET_ANALYSIS_corrVOCS MOP_MET_ANALYSIS_corrCH4 608
C2H2 MOP_MET_ANALYSIS_corrVOCS MOP_MET_ANALYSIS_corrCH4 225
HCN MOP_MET_ANALYSIS_corrVOCS MOP_MET_ANALYSIS_corrCH4 237
Aircraq observa.o
ns
q 3 experiments with ini.al condi.ons op.miza.on
Ø CH4 Ø HOx (OH + HO2) Ø VOCS
q 1 experiment with op.miza.on of the total CO emission fluxes
Ø SFCO
DA Experiments
P T U V Q CO
P
T
U
V
Q
CO
CH4
OH
HO2
VOCS
SFCO
Assimilated observa.ons
Mod
el variables
Emissions correc.on Input data : Daily gridded emissions / 1-‐ Emissions iniOal perturbaOon 2-‐ Update for t up to t +72h
Model run: Emissions allocated in one surface flux : SFCO
DART: Analysis increment in observaOons space are applied to CO and SFCO in model space
fCOmf
COm
fCOm
fSFCO
SFCO yyyx
x ,,
,
)var(),cov(Δ=Δ
fCOmy ,
xemis,ia = xemis,i
f +xSFCO,ia − xSFCO,i
f
xSFCO,if * xemis,i
f
q Seems feasible, needs to use a larger ensemble or reduced localiza.on
Emissions correc.ons
Ø Same scores for CO, reduced bias in the boundary layer when correlated with emissions
Results : comparison with NASA/Intex B DC8 flight (March 2006)
v All the 6 flights are considered, 4-‐19 March 2006 v Average in in 100hPa bin, outside Mexico city
(Emmons et al. 2010)
Met & MOPITT analysis (No cor.)
Aircraq observa.ons
Met & MOPITT analysis
Met & MOPITT analysis (cor. emissions)
Ø Met analysis Ø Forecast start on February 28th from CO analysis Ø Forecast start on February 28th from CO analysis with update emissions Ø MOPITT and Met analysis
DA Experiments: forecasts
Ø Same scores for CO, reduced bias in the boundary layer when using updated emissions
Results : comparison with NASA/Intex B DC8 flight (March 2006)
v All the 6 flights are considered, 4-‐19 March 2006 v Average in in 100hPa bin, outside Mexico city
(Emmons et al. 2010)
Met analysis
Aircraq observa.ons
Met analysis MOPITT forecast
Met analysis MOPITT forecast updated emissions
Conclusions q MOPITT CO state is correctly assimilated within CAM-‐CHEM
Ø longer analysis will be evaluated against TES CO and MOZAIC CO and O3 observaOons
q Due to its lifeOme of around 2 months
Ø Forecast iniOalize from the analysis is improve for a month Ø Forecast using updated emissions show that emissions play an important role, it is transported up to the upper troposphere
q EsOmaOon of CO emissions fluxes shows reasonable correlaOon and spaOo-‐temporal paqerns Ø Need to be evaluated in a OSSE framework Ø Will be compare with the analyOcal-‐tracer inversion
Conclusions
q Variable localizaOon ensure more stable results • Using that ensemble size / localizaOon • Hard to know if the VOCS sensibility to CO assimilaOon is going in the good direcOons
• Use tracers of CO producOon from chemistry • CorrelaOon with HOx and emissions could improve predictability for CO
q Observed correlaOons can be compare with correlaOons generated by the ensemble • Need to check significance of the evaluaOon with/of other species
Thank you ! Merci !
Ø For HCN
Results : comparison with NASA/Intex B DC8 flight (March 2006)
Met analysis
Aircraq observa.ons
Met & MOPITT analysis (no cor.)
Met & MOPITT analysis (cor. VOCS)