Assimilation of AIRS CO 2 Observations with EnKF in a Carbon-Climate - - PowerPoint PPT Presentation

1

Assimilation of AIRS CO2 Observations with EnKF in a Carbon-Climate Model

Junjie Liu, Inez Fung (UCB) Eugenia Kalnay, Ji-Sun Kang (UMD) Mous Chahine, Ed Olsen, Luke Chen (JPL)

2

=> Important to have accurate vertical mixing in the model; => Accurate 4-D (x, y, z, t) CO2 fields.

Stephens et al., 2007, (Science)

• Numbers and characters are different

transport models.

NH land

Tropical land

• bs

Vertical levels ppm

CO2

surface

• bs

Di Differ eren ent CO CO2 v vertical gradient forecasts gi give ve di differ erent ent CO CO2 f flux

3

Met Meteo eorology gy fiel elds ds & & CO CO2

Atmosphere CO2 Winds Height

• Offline transport models have been used.
• The initialization meteorology fields are from either reanalysis

products (usually 6-hourly) or off-line dynamical model;

• The vertical mixing has large uncertainty;
• Single realization of meteorological field.

4

• Generate 6-hourly 3-D (x, y, z) CO2 fields by assimilating

CO2 and meteorological observations with full GCM

Research Goals Research Goals

5

Carbon-Climate Model Carbon-Climate Model

Community Atmospheric Model (fvCAM 3.5) (2.5x1.9x26) CO2, winds, q, T, Ps Photosynthesis Respiration

Land Ocean

Fossil fuel emission Ocean CO2 flux (Takahashi et al. 2002)

• CO2 is transported as a tracer;
• Vertical mixing is updated every 30 minutes;
• Land carbon flux: 6-hourly flux from biogeochemical model.

6

Ensemble Kalman Filter (EnKF) process Ensemble Kalman Filter (EnKF) process

• Forecast error changes with time;
• Obtain ensemble analyses.

t=0hr t=06hr t=12hr

Ensemble forecasts Ensemble analyses (initial states) Observations

7

CO2 Observation Operator

• Model forecast xb is CO2 vertical profile;
• AIRS CO2 is column-weighted Volume Mixing Ratio (vmr);

=> observation operator: interpolate xb to obs location & calculate model forecast column-weighted CO2 vmr.

yb

model forecast "obs"

!

= A

avg kernel

!

T (

S

spatial interpolator

!

• bs operator

" # $$ $ % $$$ ( xb

model forecast

!

))

8

Assimilation experiments

• Met-run: assimilate raw meteorological observations (106
• bservations)
• AIRS-run: assimilate AIRS CO2 observations in conjunction

with meteorological observations.

9

The impact of AIRS CO The impact of AIRS CO2 assimilation assimilation

• n 6-hourly CO
• n 6-hourly CO2

2 3D (x, y, z) fields

D (x, y, z) fields

CAM3.5

LETKF Observations (u,v,T,q,Ps) analysis (u, v, T, q, Ps) CO2 LETKF Observations AIRS CO2 analysis (CO2) 6 hour forecast (u, v, T, q, Ps)

AIRS-run

• AIRS-run: AIRS CO2+met obs; Met-run: only met obs.
• The year of 2003.
• Prescribed surface CO2 flux forcing.

CAM3.5

LETKF 6 hour forecast (u, v, T, q, Ps) Observations (u,v,T,q,Ps) analysis (u, v, T,q, Ps) CO2

Met-run

10

• Averaging kernel: the sensitivity of AIRS CO2 to CO2

at each vertical level.

AIRS CO2 at 18Z01May2003 (+/-3hour) € €

AIRS averaging kernel

• : polar region; +:

mid-latitude; closed circles: the tropics.

Mo More e than an 2000 2000 AI AIRS CO2 w within 6 hours; mo more e sen ensitive e in the e mi middl ddle e tropo posph pher ere

ppm

11

Anal Analysi ysis s cor correct ections

• ns to
• CO

CO2 fo forecast peak at the si similar ar level evels s as as the he peak peak of

• f the

he averagi averaging ng kernel kernels

• No CO2 observations beyond 60ºS.

Time-averaged (10 months) absolute analysis corrections

Averaging Kernel

Vertical levels (hPa)

• : polar region; +:

mid-latitude; closed circles: the tropics. 90°S 90°N 0°

12

May 2003: CO2(850hPa)-CO2(400hPa) Met-run

Assimilating CO2 adjusts CO2 vertical gradient

• In the NH, CO2(850hPa)>CO2(400hPa): fossil fuel+ land carbon

source;

• In the SH, CO2(850hPa)<CO2(400hPa): transported from the NH.

13

May 2003: CO2(850hPa)-CO2(400hPa) Met-run

Assimilating CO2 adjusts CO2 vertical gradient

• In the NH, CO2(850hPa)>CO2(400hPa): fossil fuel+ land carbon

source;

• In the SH, CO2(850hPa)<CO2(400hPa): transported from the NH.

(AIRS-run)-(met-run)

• Require CO2 obs in the lower troposphere to further constrain

gradient.

14

In Inconsiste sistent sp t spatia tial d l distrib istributio tion b betw tween AI AIRS RS CO CO2 a and nd oc

• cean-air CO2 f

flux

Ocean-atmosphere CO2 flux (unit: 10-9kgC/m2/s Takahashi et al., 2002)

Annual mean AIRS CO2 spatial anomaly (ppm)

15

Tr Tropi

• pical

cal AI AIRS RS CO CO2 r relates to ci circulation and and aver averagi aging ng ker kernel nel

Ocean-atmosphere CO2 flux (unit: 10-9kgC/m2/s Takahashi et al., 2002)

Annual mean AIRS CO2 spatial anomaly (ppm)

16

Annual mean CO2 correction from assimilating AIRS CO2

As Assimi milat ating g AI AIRS RS CO CO2 i improves spatial pat patter ern

Annual mean AIRS CO2 spatial anomaly (ppm) CO2 spatial anomaly at AIRS CO2 space from met-run

17

Co Consisten ent CO CO2 d distribution and we weather pat patter ern

Single time (12Z27Feb2003) Time average over Feb 2003 ppm 500hPa geopotential height (contour) and CO2 from AIRS-run (shaded)

• Simultaneous assimilation of meteorology variables and CO2

generates CO2 distribution consistent with weather pattern

18

CO CO2 a analysis spread ran ranges ges from 0. 0.4ppm 4ppm to

• 2ppm

2ppm at at 400hPa 400hPa

400hPa monthly mean (September) CO2 spread

• Analysis ensemble spread is related to observation coverage,

forecast error and observation error;

• Larger spread over high latitudes, and over land;
• Smaller spread over tropical ocean is due to observation coverage

and propagation through forecast. ppm Average num of CO2 observations at each grid box within 6 hours

19

Co Column mn-integ egrat ated ed CO CO2—Se Sep

AIRS AIRS

• run

Met- run AIRS-run - AIRS Met-run - AIRS ppm

ppm

20

Co Column mn-integ egrat ated ed CO CO2—Oc Oct

AIRS AIRS

• run

Met- run AIRS-run - AIRS Met-run - AIRS ppm

ppm

21

Co Column mn-integ egrat ated ed CO CO2—No Nov

AIRS AIRS

• run

Met- run AIRS-run - AIRS Met-run - AIRS ppm

ppm

20 days with no AIRS CO2

22

Assi Assimilat ating ng AI AIRS RS CO CO2

2 im

improves surface CO CO2 s seasonal c l cycle le a and th the N- N-S gradient

Surface obs: black; Met-run: red: AIRS-run: blue

Mean NH CO2 concentration at 8 surface stations The N-S gradient based on 16 surface stations

Surface data is from NOAA/ESRL website

Met-run has similar NH CO2 concentration and the N-S gradient

23

Assi Assimilat ating ng AI AIRS RS CO CO2

2 im

improves CO2 state state esti estimate ate

• CO2 from the AIRS-run can be about 1 ppm more accurate than

those from the met-run. Height (km)

1km 8.5km rms error (ppm) Briggsdale, US Molokai Island

Verified against independent aircraft CO2 observations

Cook Islands Met-run: no CO2 obs

AIRS-run: assimilate CO2

24

Prel Prelimi minary nary resul results s on

• n surf

surface ace carbon carbon flux ux est estimat mation

• n by

by assi assimi milat ating ng AI AIRS RS CO2 CO2

25

The impact of AIRS CO The impact of AIRS CO2 assimilation assimilation

• n surface CO
• n surface CO2 flux

flux

CAM3.5

LETKF Observations (u,v,T,q,Ps) analysis (u, v, T, Ps) (CO2) LETKF AIRS CO2 and conventional CO2

• bservations

analysis (CO2 Cflux) LETKF: Local Ensemble Transform Kalman Filter (Hunt et al., 2007)

• The carbon flux analysis acts as boundary forcing for the forecast
• f next time step.
• Four and a half months assimilation cycles (01Jan2003-10May2003).

6 hour forecast (u, v, T, q, Ps)

26

Car Carbon Flux An Anal alysis:Dat Data a As Assim (l (left) Car Carbon Flux (CAS CASA A (lan and)+Tak akah ahas ashi (ocean cean))(right)

January 2003

• AIRS has the most impact over the tropical land
• Stronger source in the NH winter
• Stronger sink in the tropics and SH subtropics
• Noisy over ocean compared to Takahashi

10-8 kg/m2/s

27

Car Carbon Flux An Anal alysis:Dat Data a As Assim (l (left) Car Carbon Flux (CAS CASA A (lan and)+Tak akah ahas ashi (ocean cean))(right)

February 2003

• Stronger source in the NH winter
• Stronger sink in the tropics and SH subtropics
• Noisy over ocean compared to Takahashi

10-8 kg/m2/s

28

Car Carbon Flux An Anal alysis:Dat Data a As Assim (l (left) Car Carbon Flux (CAS CASA A (lan and)+Tak akah ahas ashi (ocean cean))(right)

March 2003

• Little change

10-8 kg/m2/s

29

April 2003

• Little change

Car Carbon Flux An Anal alysis:Dat Data a As Assim (l (left) Car Carbon Flux (CAS CASA A (lan and)+Tak akah ahas ashi (ocean cean))(right)

10-8 kg/m2/s

30

May 2003

• As in the OSSEs, the surface fluxes appear initially

to be reasonable and then they “get stuck”.

Car Carbon Flux An Anal alysis:Dat Data a As Assim (l (left) Car Carbon Flux (CAS CASA A (lan and)+Tak akah ahas ashi (ocean cean))(right)

10-8 kg/m2/s

31

Monthly average AIRS CO Monthly average AIRS CO2 does not does not change hange much much over the tropical land from January to May ver the tropical land from January to May

Jan Feb Mar April

32

Conclusions - General

• EnKF brings important advantages for Reanalysis:

 Analysis uncertainty  Adaptation to new observing systems  Estimation of obs. errors and identification of bad

• bservations (not shown)

 Estimation of model bias (essential)

• For Carbon Reanalysis it is essential to assimilate at the

same time meteorological and carbon observations

33

Conclusions - Simulations (OSSEs)

• The advantage of the OSSEs is that we know the true

fluxes and CO2

• It is possible to estimate surface carbon fluxes from

atmospheric CO2 measurements but  Need “variable localization” to reduce sampling errors  Need adaptive inflation of the B error covariance  Need to estimate model bias

• Problem: the initial results after spinup from random

fields are good, but then the surface fluxes “don’t want to change anymore”.  This is probably due to model bias

34

Conclusions - AIRS data assimilation

• AIRS CO2 data assimilation is clearly successful!

 Improved atmospheric CO2 and N-S gradient  Better agreement with independent observations  Insight about vertical circulation and mixing

• Preliminary estimations of carbon fluxes are very

promising after one month spin-up:  Compared with CASA fluxes they yield reasonable uptakes in the SH summer and stronger emissions in the NH winter  But, like in the OSSEs, the fluxes “don’t want to change” with season

35

Conclusions - AIRS/IASI GoSAT/OCO2

• The combination of satellite and in situ data is important:

 Results are more accurate in NH than in SH

• We need more near surface information

 Carbon fluxes can be derived from atmospheric CO2

• Results depend on optimal forecast spread, a difficult

problem for surface fluxes:  Work on estimating model bias  Should find why after initial good surface fluxes they “don’t want to change” with season  We probably need to estimate diurnal and seasonal changes with a different approach (e.g., EOFs).