GERB-2 V999 Unfiltering

Nicolas Clerbaux & RMIB GERB Team.

GIST 24, Imperial College, London

15 december 2005

Outline

• Shortwave– V999 spectral response – V999 unfiltering

• New : Empirical SEVIRI NB-2-BB• New : Clear ocean unfiltering

– Validations/main results• Clear desert unfiltering problem• Clear ocean NB-to-BB problem

– Solution for release

• Longwave– V999 spectral response– V999 Unfiltering– Unfiltered radiance comparison with CERES

Comparison V002/V999 SW spectral response curves

• V999 (green) seems flatter but...

• ... unfiltering more difficult

V999 SW Unfiltering

• Same unfiltering method as before:

• New empirical BB estimation with parameters dependent on surface type (CERES ST used: ocean, dark vegetation, …):

-> Unfiltering parameters are dependent on surface type• Unfiltering parameterization based on a new data base of

spectral radiance curves (SBDART 2.4)• Implementation of a dedicated clear ocean unfiltering (to

try to solve the asymmetry problem).

filnb2bb

unfilnb2bb

LLα

01.60.80.6BB θ eρ dρ cρ baρ

• Database of 3 months (March, April and July 2004) of coangular SEVIRI NB and CERES BB radiances (FM2 and FM3), (more than 4 millions observations),

• Derive regressions in terms of reflectance, for V999:

• Further improvement presented during the EUMETSAT conference at Dubrovnik (more info in the proceedings)

New: Empirical SEVIRI NB-2-BB

β gρ fθ eρ dρ cρ baρ 20.601.60.80.6BB

01.60.80.6BB θ eρ dρ cρ baρ

sun glint angle

New: Dedicated clear ocean unfiltering

• As different as possible from the previous version to see if the asymmetry still there

• Direct estimation of the unfiltering factor as a function of

0.6:

=a+b/0.6

+c/0.6

with a,b,c dependent on

().

• Best fit on the ocean scenes (clear or cloudy) in the database.

SW Unfiltering Validations

Analysis of– Dave Doelling– Richard Allan– Cedric Bertrand– Steven Dewitte

Main results:– Clear desert overestimation,– “Diurnal cycle” of error for the SEVIRI BB estimate

(GERB-like), specially over the clear ocean,– This error is only partly corrected by GERB in the

high resolution products (ARCH, SHI),– Dedicated clear ocean unfiltering not needed (direct

unfiltering seems sufficient)– Diurnal cycle of unfiltering error for dark scene?

Clear desert problem

• V999 clear desert overestimation wrt to the other scene types (about 2%?)

• Reason: the desert unfiltering was parameterized using the “Soils” ASTER library category and not the “Rocks”

• Preliminary simulations using “Rocks” explain about 1.5% of the overestimation

• Solution: add new simulations in the data base and re-parameterize the unfiltering for the desert surfaces

• See presentation of Cedric Bertrand. • Possible solutions:

– Do not use NB-2-BB but well the CERES TRMM ADM radiance,

– Add to the NB-2-BB a correction c(SZA,VZA,RAA) dependent on the full angular geometry (parameterization using RTM),

– Improve the NB-2-BB (more proxi) • Validations

– Comparison diurnal cycle with model– Data base of 132 695 coangular CERES and SEVIRI clear

ocean observations: L_ceres = a L_nb2bb

Clear Ocean NB-2-BB Problem

Method Slope(a) RMS error (%)

Radiance from ADM 0.982 18.2%

V999 NB2BB 0.974 16.7%

V999 NB2BB + Corr 1.016 13.0%

Dubrovnik NB2BB 1.004 14.1%

Dubrovnik NB2BB + Corr 0.991 12.9%

See RMIB TN: “Clear ocean correction for SEVIRI NB-to-BB Conversion”

• Select/implement on method to improve clear ocean BB estimation

• SBDART simulations with ASTER rock category + check the clear desert improvement

• Additional unfiltering validations needed for dark scenes (dark vegetation- tropical forest). If needed, use direct unfiltering instead of the ratio of the BB estimations

To do before release - SW

Comparison V002/V999 LW spectral response curves

• V999 more sensitive to cold scene

• Bug in the V002 solar LW contamination

• Spectral response extended up to 500µm

LW Unfiltering

• Same as before except that the unfiltering factor is now estimated over [2.5:500]µm instead of [2.5:100]µm for V002

Validation with CERES FM3 & FM2

• V999 ARCH data for June, July & December 2004.

• Analysis of the averaged coangular LW radiance:– at regional scale (10*10 ARCH pixel

boxes)– in radiance bin (5Wm-2sr-1)

See RMIB TN: “Cross validation of GERB LW radiance with CERES”

FM3 daytime June/July 2004

<Lceres> [Wm-2sr-1] ratio : <Lgerb>/<Lceres>

FM3 nighttime June/July 2004

<Lceres> [Wm-2sr-1] ratio : <Lgerb>/<Lceres>

FM3 June/July

2004

FM2 daytime June/July 2004

<Lceres> [Wm-2sr-1] ratio : <Lgerb>/<Lceres>

FM2 night time June/July 2004

<Lceres> [Wm-2sr-1] ratio : <Lgerb>/<Lceres>

FM2 June/July

2004

Discussion- LW

• about 2% scene type dependency

• A bit higher for V999 than for V002

• Should be confirmed with ARG data

• Over daytime hot desert ratio GERB/FM2 not consistent with GERB/FM3. Reason?

Thank you !