Eumetsat GOME-2 Error Assessment Study

Interim Findings

Presentation by B.Kerridge

on behalf of Serco, RAL, IUP & SRON

GSAG, ESRIN, 11/12th April 2002

Structure

1 Introduction2 Approach3 Baseline Error Budget4 Interim Findings:

(a) Sampling Options for Band 1

(b) Spatial Aliassing

(c) Spectral Resolution & Slit-Function Shape

(d) RTM Assumptions & Earth Curvature

(e) Non-Lambertian Surface BRDF

(f) Cloud Obscuration & Horizontal RI Gradients

(g) Pointing & Geolocation4 Summary & Further Work

1. Introduction

• GOME-2 Error Assessment Study commissioned by Eumetsat• Scope:

– Identify through quantitative retrieval simulations factors which will limit accuracies of trace gas columns and ozone profiles.

– Recommend operational settings and, if necessary, other action to mitigate these.

• Consortium:– Serco Europe Ltd (Prime Contractor)

– RAL (Technical Co-ordinator, Ozone Profile Analysis)

– IUP (Trace Gas Column Analysis, RTM Calculations)

– SRON (Assessment of Instrumental Errors)

• Final Presentation:– 25th June at Eumetsat

2. Approach

• Trace gas column / O3 profile algorithms as used by IUP / RAL for GOME-1 flight data, with specific modifications for GOME-2.

• For O3 profiles, sensitivity to -ranges and a priori explored.• Simulations generally for:

• 12 geo-temporal scenarios with realistic tropospheric and stratospheric trace gas profiles and solar geometry from orbit propagator

• 2 surface albedos (0.05, 0.8)• Variety of view angles (nadir, +/- 29o, +/- 45o (=1920km swath))

• Linear retrieval diagnostics analysed.• Specified errors for GOME-2 quantified by linear mapping of

their spectral signatures for comparison with:(a) Estimated Standard Deviations (ESD = sqrt(Sx) )

(b) Baseline Error Budget

3. Baseline Error Budget

Baseline error budgets based on GOME1 experience and using GOME2 noise model

O3 Profile Errors: – Radiometry: 2% of sun-normalised radiance– Polarisation correction: SRON prescription for GOME-2– Degradation of scan-mirror reflectance: quantified but neglected – Surface pressure: 10hPa– Temperature profile: error covariance matrix from IASI retrieval– Aerosol: LOWTRAN “high” - “background”

Trace Gas Column Errors:– Photon noise & read-out noise, with Ring x-section fitted.– Polarisation correction: error quantified but negligible.– Largest error source caused by differential structures from diffuser plate

reflectivity, errors in trace gas columns exceeding >50% except for ozone (see Richter and Wagner, 2001)

• Clouds, spectroscopy and other instrument effects not included.

4. Interim Findings

(a) Sampling Options for Band 1• Baseline integration times: 12s for Band 1A; 0.1875s for Band 1B Single Band 1A pixel: 1920km x 80km

– Non-linear dependence of RT with view angle over +/- 45o

– RT errors at extreme view angles

– Horizontal variability of stratospheric O3 profile

• Flexibility to read out Band 1A at 1.5s (640km x 40km) and co-add• Impact of increased read-out noise assessed, in frame of 1% noise

floor• Co-adding 1.5s Band 1A pixels to 12s (640km x 320km) or 24s

(640km x 640km) yields similar esd to single 12s Band 1A pixel.• Additional read-out noise insignificant also for co-addition of

0.1875s Band 1B pixels No impediment to reading out Band 1A at 1.5s & Band 1B at

0.1875s

(b) Spatial Aliasing• Five Landsat ETM+ images (~180km x 180km at 1m resolution).• Ensemble of >350 spatially-aliassed signatures calculated (via

dependent surface albedos) for each of the12 geo-temporal scenarios.

• Ensemble min, max, mean (bias) and RMS examined.• GOME-2 IFOV (0.29o ~4km on ground) effectively filters high-

frequency structure (ie spatially-aliassed noise). Coarser resolution images (eg ATSR-2 1km x 1km) would suffice for a

future comprehensive study, and are calibrated over full dynamic range.

Trace Gas Columns• Fitting windows small, so errors insensitive to low-frequency

structure.

• Max errors (0.02% O3, 2% NO2, 1% BrO, 10% OClO); below the esds No reason to reduce detector read-out time.

O3 Profiles• Bands 1 & 2 handled as separate steps.• Band 1 sees low-frequency structure only at longest wavelengths.• Errors <esd for Band 1 range 265-307nm, but sometimes exceed esd

when Band 1 range extended to 265-314nm (nb albedo extremes). Reduction in Band 1B detector read-out time desirable. • Errors arise from non-linear RT dependence on surface albedo plus

scene inhomogeneity (ie not eliminated for negligible read-out time). • Only small interval (~100 of 1024 detector pixels) of Band 2 used,

and 2nd order polynomial fitted to log(sun-normalised radiance). Band 2 retrieval insensitive to low-frequency structure from

aliassing. • Algorithms which use (a) sun-normalised radiances and (b) extensive

intervals of Bands 1 and 2 simultaneously would be more vulnerable.

(c) Spectral Resolution & Slit-Function Shape

O3 Profiles

Slit-function FWHM:• Trade-off between increasing esd and decreasing undersampling

error as Band 2 FWHM is increased (redistributing but conserving photons).

• On this criterion, increase from 2-3px (0.24-0.36nm) would be beneficial, however knowledge of shape would then be more critical.

Slit-function Shape:• Gaussian shape assumed in Band 2 FM for joint retrieval of:

– Slit FWHM and wavelength registration from direct-sun spectra

– O3 and other variables from sun-normalised radiance spectra

• Analysis (OG) for EQM indicates non-Gaussian and asymmetric shape

• Shape for defocused case broader and more wavelength dependent, although closer to Gaussian, than for focused case.

• Measurements of direct-sun and sun-normalised spectra synthesised for “true” shapes (focused, defocused, Gaussian) using high-res solar spectrum, adding Ring to backscattered spectrum before convolution.

• Gaussian used in retrieval scheme FM.• Two simulations w.r.t. undersampling:

-shift (direct-sun - backscattered) retrieved (as GOME-1) -shift not retrieved but mapped, after interpolation & re-

gridding.

• Errors on retrieved O3 small when “true” shape really is Gaussian (ie FHWM of Gaussian recovered by retrieval), but large when it is not.

• Errors larger for defocus (wider) case than for focus case – because, although smaller in amplitude, the spectral signature due to

erroneous shape is more correlated with structure of O3 Huggins Band.

• O3 errors due to incorrect shape can exceed 100% in troposphere.

• Sign, magnitude and height-dependence of this error are similar to bias found for GOME-1 (RAL data) from comparison with >2,000 sondes.

• Characterisation of shape with onboard line lamp will be limited by:– Absence of suitable lamp lines between 306 and 333nm – Discrete sampling of narrow lines by broad detector pixels

Characterisation of Band 2 -dependent slit-function shape at sub-px resolution vital for accurate GOME-2 tropospheric O3 retrieval

Trace Gas Columns

• High res FTS absorption x-sections used.• Simulation of FWHM increase due to defocusing from 2px to

5px (0.24-0.6nm Band 2 & 0.5-1.25nm Band 3). Esds for O3 (<1%) and BrO (<60%) increased by only factor

~1.1. • NO2 esd (<20% at 2px) increased by factor ~1.24 at 3px.• Undersampling errors small for O3 (<0.5%) and NO2 (<2%),

but substantial for BrO (<100%). • Assessment of slit opening (increased photon flux):

– Error improvement by 30 percent by doubling slit width– Desirable to reduce integration time <0.1875s, hence ground

pixel size, to avoid possible saturation Slit opening improves SNR and avoids undersampling

problems

(d) RTM Assumptions & Earth Curvature• Assumption: fully-spherical RTM too computationally expensive

for operational processing.• Calculations by CDI RTM in pseudo-spherical and fully-spherical

modes differenced and linearly-mapped.

O3 Profiles• CDI (with GOME FM x-sections) also used to calculate K’s. • Pseudo-spherical approximation (as implemented in CDI) can

cause substantial errors at all altitudes (ie in ss- as well as ms-domain).

• Largest errors at largest SZAs, as expected.• Errors small for nadir-view, and much larger at +/-45o than at +/-

29o

Correction scheme required for outer pixels of 1920km swath• Caveat: errors estimated from non-linear simulations might be

smaller.

Trace Gas Columns• Errors on O3, NO2 and BrO slant-columns negligible (<1%),

provided solar geometry for ground rather than TOA is used in plan-parallel atmosphere

• NOTE: RTM errors will end up as AMF error rather than slant column errors

Does not require hardware changes, improvements possible by using most appropriate RTM assumptions compromising processing speed and accuracy

(e) Non-Lambertian Surface BRDF• Spectra calculated by CDI in fully-spherical mode using angular-

dependent surface BRDFs and their Lambertian equivalents for:– dark land, bright land, ocean & snow (April 55oN) & sunglint (5oS)

O3 Profiles• Quasi non-linear simulation: gross deviations in BRDF

accommodated through surface albedo retrieval, as per GOME-1 scheme.

• O3 errors from Lambertian assumption <5%, except >30% for sunglint.

• Sunglint occurs in eastward views, peaking near 960km swath edge.

• It will affect a substantial fraction of data in tropics south of equator. (Peak intensity and affected area depend on surface wind-speed.)

Trace Gas Columns• Difference in slant column fits from BRDF and from semi-

hemispherically integrated albedo provides error estimate

• Errors on O3, NO2 and BrO slant columns for most surface types <1%

• Errors due to sun-glint can be up to 3% for all trace gases• NOTE: This is again an AMF error rather than slant column

error

Small error, improvement possible by retrieving Lambertian equivalent albedo directly from the spectra

(f) Cloud Obscuration & Horizontal RI Gradients• Issue: extent to which cloud obscuration or horizontal gradients

in refractive index would be more serious for 1920km than 960km swath.

Cloud:• ATSR-2 forward view (55o) ~ GOME-2 extreme (45o) for 1920km • ATSR-2 statistics analysed for GOME-1 80x40km px for one year Forward/nadir differences not substantial, even for occurrence

of totally cloud-free scenes (12% vs 14%).Horizontal Gradient in RI:• LOS path-lengths calculated for non-refracting and refracting

(w/wo 0.14K/km gradient 0-40km height) atmosphere with ray-tracing model

Differences negligible for extreme view angle for 1920km.

(g) Pointing & Geolocation• Errors as built taken from EPS Geolocation & Co-registration

Budget– Nadir: 1.6km along-track, 1.2km across-track– 1920km swath edge: 3.1km along-track, 3.6km across-track

• Direct impact on viewing geometry quantified

Trace Gas Columns• Errors <1% and generally ~0.5%

O3 Profiles• Across-track errors generally <2% (even at 1920km swath

edge)• Along-track errors <<1%

Direct impact of pointing errors at these levels is negligible cf others.

5. Summary and Further Work

1. Band 1 Sampling: – Integrate for 1.5s (1A) / 0.1875s (1B) and co-add; retain 1A/1B

boundary.

2. Spatial Aliassing: – Errors on trace gas columns not significant – Errors on O3 profiles ~ esds for Band 1 limit of 307nm, > esds for

314nm. Reduction in read-out time desirable.

3. Pointing Errors: negligible direct impact 4. Spectral Resolution & Slit-Function Shape:

– FWHM increase from 2-3px would reduce sensitivity to undersampling, but would increase sensitivity to errors in knowledge of shape.

– Accurate knowledge of shape in Band 2 <350nm vital for O3 profiles. Pre-flight measurements required, since onboard line-lamp not

adequate.– Opening slit permits smaller ground pixel or improved SNR and

avoids undersampling

5. Swath-width:– Cloud obscuration & horizontal RI gradients not significant

factors.– Errors from pseudo-spherical approximation negligible for trace

gas columns but large for O3 profiles in outer pixels of 1920km swath.

Correction scheme needed if CPU time too great for fully-spherical RTM. – Errors from Lambertian BRDF approximation negligible for trace

gas columns and <5% for O3 profiles except for sunglint, where they are large. Sunglint more pervasive for 1920km swath, but not decisively.

Further study required to:(a) Address identified issues in greater depth (eg swath, slit-shape).(b) Address issues not covered by this study (eg spectroscopy,

diffuser spectral structures, DOAS/AMF assumptions, possible improvements in hardware for 3rd generation GOME).