On the biases in AIRS retrieval of ozone (work in progress) AIRS - - PowerPoint PPT Presentation

1

On the biases in AIRS retrieval of ozone

(work in progress)

Bill Irion, Michael Gunson - Jet Propulsion Laboratory Michael Newchurch - U. Alabama at Huntsville Sunmi Na - Pusan National University

With thanks to Sung-Yung Lee, Bob Oliphant, John Worden, John Blaisdell, Chris Barnet and SHADOZ

AIRS Science Team Meeting - March 9, 2006

2

AIRS captures UTLS ozone events

May 21/2005 270 mb

Filled dots are TES observations

100 600 O3 vmr (pptv) 300

AIRS TES

(filled dots)

(AIRS - TES) / TES (%)

AIRS-TES relative difference

10-8 10-7 10-6 TES O3 VMR

• 100

300

• 100

300

AIRS in qualitative agreement with TES in ozone regions > 100ppb.

3

Simplified AIRS retrieval of ozone Regression Cloud Clearing

(Constrained)

Physical Retrieval

First guess Cloud- cleared radiances and errors

Ozone profiles and columns

Damping parameter

(“noise propagation threshold”)

Training by ECMWF ozone Cloud-cleared radiances L1B AIRS 3x3 golfball obs. + AMSU Channel selection

4

How much is AIRS getting its skill in ozone from regression? …biases are similar to ECMWF.

Like ECMWF, AIRS is too high in troposphere and too low in stratosphere; column OK. AIRS/Sonde

Correction: matchup is within 100 km and 6 hrs.

ECMWF/Sonde Relative differences of AIRS & ECMWF vs ozonesondes

Standard deviations shown. Pressures

• ffset for

clarity.

5

How does channel selection and damping affect the retrieval? Regression Cloud Clearing

(Constrained)

Physical Retrieval

First guess Cloud- cleared radiances and errors

Ozone profiles and columns

Damping parameter

(“noise propagation threshold”)

Training by ECMWF ozone Cloud-cleared radiances L1B AIRS 3x3 golfball obs. + AMSU Channel selection

6

Decreasing damping worsens results in upper trop/lower strat with current channel selection

Current damping Less damping Matchups within 100 km and 3 hrs of sonde launch

Average

AIRS - Sonde Sonde

Error bars are std. dev.

• 1

2

• 1

2

7

If results worse with decreased damping, let’s give the retrieval more information

ad hoc selection O3, CO2 and H2O line strengths, frequencies and O3 retrieval channels

8

Adding channels at current damping doesn’t change anything.

Average

AIRS - Sonde Sonde

Error bars are std. dev.

9

Adding channels and decreasing damping gives mixed results

Current damping & current channels Less damping and more channels Worse Worse Worse Better Better Better Better Better Better

Point: there’s some tradespace with decreased damping and additional channels.

Average

AIRS - Sonde Sonde

Error bars are std. dev. Pressures offset for clarity

10

Let’s look at the radiances and their uncertainties… Regression Cloud Clearing

(Constrained)

Physical Retrieval

First guess Cloud- cleared radiances and errors

Ozone profiles and columns

Damping parameter

(“noise propagation threshold”)

Training by ECMWF ozone Cloud-cleared radiances L1B AIRS 3x3 golfball obs. + AMSU Channel selection

11

We need reliable errors in cloud-cleared radiances!

= 1 N

• bsi calci

NESRi

• 2

i=1 N

• If χ >> 1, bad fits or underestimating noise

If χ << 1, fitting noise or overestimating noise

χ= 0.28

Error too high?

χ= 1.03

Error about right?

χ= 2.33

Error too low?

Red line = radiance error in obs BT for O3 channels only Black marks = (obs - calc) in BT

^2

9/6/2002 granule 176

geoTrack 2, geoXTrack 27 geoTrack 3, geoXTrack 23 geoTrack 21, geoXTrack 5

• 10

10

• 1

1

• 1

1

Overly high error in cloud-cleared spectral radiance helps drive over-constraint of retrieval. Overly low errors help drive an under-constraint. Goodness of fit diagnostic

Cloud-cleared radiance error

12

Systematic biases in radiance uncertainties?

Mean BT error in ozone channels (K)

Qual_O3 = 0 Qual_surf = 0

= 1 N

• bsi calci

NESRi

• 2

i=1 N

• If χ >> 1, bad fits or

underestimating noise If χ << 1, fitting noise or

• verestimating noise

Sept 6/02 V4 Granule 176

(mostly ocean off US Northeast) 1 10 100 0.1 1 10 100 0.1 0.01 0.01 100 10 0.01

χ vs mean BT error in ozone channels

The biggest problem with ozone may not be in the regression or the physical retrieval, but in the cloud-clearing.

χ vs BT error should be a horizontal line at χ=1 !

χ

13

Summary

Regression Cloud Clearing Physical Retrieval

ECMWF O3 biases in first guess Incorrect uncertainties in 10 µm band radiances

Ozone profile biases

Damping parameter set too tightly? Retrieval

• ften too

constrained

Work in progress. Note that for the moment I’m not taking into account trapezoids, biases in the spectroscopy, etc.

Suboptimal channel selection?

14

Thanks for your time!

Retrieval Damping Squad

15

Surface Temperature

Qual_surf = 0

χ

Mean BT error in fitted channels (K)

16

Temperature Profile

Qual_surf = 0

χ

Mean BT error in fitted channels (K)

17

χ vs Brightness Temperature Error Optimal Estimation Retrieval

= 1 N

• bsi calci

NESRi

• 2

i=1 N

• If χ >> 1, bad fits or underestimating noise

If χ << 1, fitting noise or overestimating noise