Data assimilation of trace gases in a regional chemical transport model: the impact on model forecasts E. Emili 1, O. Pannekoucke 1,2, E. Jaumouillé 2,

Data assimilation of trace gases in a regional chemical transport model:

the impact on model forecasts

E. Emili1, O. Pannekoucke1,2, E. Jaumouillé2, A. Piacentini1, D. Cariolle1 & Météo-France MOCAGE

team

1Centre Européen de Recherche et de Formation Avancée en Calcul Scientifique Toulouse, France

1

2

Overview 2

• Description of the regional assimilation system

• Observation and background errors

• Assimilation of surface O3, NO2 and CO

• Skills of short-term forecasts

• Assimilation of satellite O3 and NO2

• Conclusions

MOCAGE + variational data assimilation = VALENTINA 3

o Semi-lagrangian chemical transport model: MOCAGE (Météo France - CNRM)• Global (2.0 deg) and regional (0.2 deg) nested domains• 47 vertical levels up to 5 hPa• RACM+REPROBUS chemistry (91 transported species, >300 reactions)• IFS (ECMWF) meteorological forcing• TNO+MACCity emissions

Surface model increments of O3 (ppbv) after the assimilation of European in-situ data (AIRBASE):

o Variational data assimilation: VALENTINA (CERFACS)• 3D-Var (1h assimilation windows)• Initial state optimization • Assimilation of column/profile/surface

data• 3D background error matrix (B)

o Operations within MACC-II project:• Daily forecasts (96h)• Daily re-analyses (D-1)• Yearly re-analyses (2008-2012)

Summer-time O3 episode: 6-12 July 2010 4

6-7-2010 12 UTC 8-7-2010 12 UTC 10-7-2010 12 UTC 12-7-2010 12 UTC

Surf

ace

tem

pera

ture

(C°)

Surf

ace

O3 (

ppbv

)M

easu

red

O3 (

ppbv

)

AIRBASE O3 hourly measurements (≈1000 sites):

MOCAGE control simulation:

Meteorological forcing (IFS):

EU O3 limit

Observations and background errors 5

O 3 super-observations averageStandard deviation of measurements within 50 km

1) Selection of super-sites with n>4 sites within 50 km (twice the model grid resolution)

2) Calculation of spatial standard deviation (o) for each super-observation-> temporal average

3) Calculation of model Rmse (control simulation – super-observations) at each super-site

4) Background error B = (Rmse2-o2)1/2 -> sites

average

B raw approximation of background error: overestimation of true forecast error, missing temporal variability, biases are counted in the error. Background horizontal error correlation fixed to 25/12 km for O3,CO/NO2

Ozone observation error o in July 2010 (ppbv)

Ozone background error B in July 2010 (ppbv)

Observations and background errors 6

July 2010

January 2012

Average Observation error (%)Average Background error (%)

• NO2 observation and background errors are the highest (small scale variability, emissions uncertainty, vertical mixing…)• O3 observation error is about 1/3 of background error• CO observation and background errors are comparables• O3 background error doubles in winter (under investigation)

7Analysis skills• Control run without data assimilation• Analysis• Independent observations

• Control run without data assimilation• Analysis• Independent observations

8Forecasts skillsJanuary O3 July O3

January NO2 July NO2

January CO July CO

Diurnal cycle of model and observations values:• Control run without data assimilation• 00+24h possible MACC forecast (initialized with the

analysis at 00 UTC)• 1h forecast, e.g. the assimilation background• Analysis• Independent observations

Forecasts skills 9

MACC analysis for D-1 produced at 09 UTC (observations collection): delay between D forecast and latest analysis > 9h

Which gain if this delay could be reduced?Rmse %Bias %

Forecasts skills 10

What is the impact of assimilating one species on the forecast of the others?Validation of O3 1h forecasts:

7-17 Jul 2010

15-25 Jan 2012

11

Validation of IASI (MetOP) and surface analyses with ozonesonde data (6-12/7/2010):

OzonesondesMOCAGE control simul.

hPa

Number of ozonesondes profiles available during the episode (14 totally, 11 in continental Europe)

• IASI corrects the free troposphere O3 negative bias

• Surface obs correct the positive bias at the surface

• The combination of the two gives the best model profile up to 200 hPa

OzonesondesSurface obs. analysis

OzonesondesAIRBASE+IASI analysis

OzonesondesIASI analysis

hPa

34 2

11

Negative free troposphere bias

Positive surface bias

Satellite and surface O3 analysis

12

NO2 tropospheric columns from OMI (Aura) and GOME-2 (MetOP) assimilated in MOCAGE:- Surface bias reduction (10%) at rural sites in winter- Not significant impact observed in summer (too short NO2 lifetime)

Control run columns Satellite columns Analysis columns

Satellite NO2 analysis

Conclusions and perspectives 13

European surface re-analyses well constrained by the dense AIRBASE network. O3 re-analysis scores better than NO2 and CO ones, what is the impact of observation error in validation exercises?

Forecasts skills depend on the species and the season. Reducing the delay from the last available analysis to 3h might reduce the forecast bias by a factor 3 (O3 in summer).

A positive impact of correcting one species on the forecast of other species is not demonstrated. Need deeper investigation of the chemical system.

Assimilation of satellite data corrects model tropospheric columns, but positive impact at the surface is not clearly demonstrated. Need satellite products with enhanced boundary layer sensitivity.

14

Thanks for your attention

Centre Européen de Recherche et de Formation Avancée en Calcul Scientifique Toulouse, France