CHEOPS-GOME Study on Seasonal effects on the ERS-2/GOME ... · In channel 2, seasonal dependence in the residuals is not apparent, but here the picture is dominated by jumps in the

Deutsches Zentrumfür Luft- und Raumfahrt e.V.DLR

CHEOPS-GOME

Study on Seasonal effectson the ERS-2/GOME Diffuser BSDF

Author: Sander SlijkhuisDoc.No.: CH-TN-DLR-GO-0001Issue: 1Date: 9 May 2004

Deutsches Zentrum für Luft- und Raumfahrt e.V. - DLRInstitut für Methoden der Fernerkundung - IMF

OberpfaffenhofenGermany

CHEOPS-GOME: BSDF seasonal effectsCH-TN-DLR-GO-0001Issue 19 May 2004

2

Table of Contents

1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31.1 Study rationale and basic principles . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31.2 Results of earlier work . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31.3 Outline of this study . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4

2 Variability of solar irradiance 1995-2002 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 53 Seasonal dependence of the ‘smoothed’ BSDF. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7

3.1 Method . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 73.2 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8

4 Seasonal dependence of the BSDF without wavelength smoothing . . . . . . . . . . . . . . . . . 145 Discussion. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22

5.1 Summary of results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 225.2 Reliability of the results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 225.3 Relation to Ozone profile retrieval . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24

6 Reference Documents . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 257 Annex A: Relevant figures based on GDAQI work . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26

CHEOPS-GOME: BSDF seasonal effectsCH-TN-DLR-GO-0001

Issue 19 May 2004DLR

3

lcula-dianceSolarciple,er -

ed at adue torther-

nationr if

rome-

sure-adi-mentsnctionorbit.

of

thef the-prod-lyno-

f thisntly

howsere asdeg-

als.nnel 2,

y jumpsally inthe iceportgth thele sea-

ration

1 Introduction

1.1 Study rationale and basic principles

Trace-gas profile retrievals from the GOME instrument are made by comparing model cations of atmospheric transmission and backscattering to the observed ratio of Earthshine Rato incident Irradiance. In the GOME instrument, the latter is observed once a day by makingmeasurements with a diffuser in the lightpath. The ratio Radiance/Irradiance is thus, in prinonly determined by the Bi-directional Scattering Distribution Function (BSDF) of the diffusan absolute radiance calibration of the instrument is not necessary.

In practice, the situation may be more complex: the Irradiance measurements are performdifferent scan mirror angle than the Radiance measurements, which may invoke artefactsthe instrument’s polarisation sensitivity, or due to scan-angle dependent degradation. Fumore, it has been observed that changing interference patterns arise with changing illumiconditions of the diffuser [R1][R2] - it is unclear if these features arise only from the diffuser othey are an interplay between beam-filling effects and interference layers in the main spectter.

In this study of BSDF seasonal dependence, we will use the variablility in the irradiance meaments (corrected for Sun-Earth distance variations) as a proxy for variability in the ratio Rance/Irradiance. There is no a priori reason why the optical path in the Radiance measureshould change with season: although the incident solar illumination varies over season as fuof sub-satellite latitude, nearly all illumination conditions keep occuring somewhere on an

This study is carried out in the framework of the CHEOPS-GOME project: “ClimatologyHeight-resolved Earth Ozone and Profiling Systems for GOME”.

1.2 Results of earlier work

As basis of this study we use work performed, and software written, by E. Hegels (DLR) inframework of the GDAQI study [R3]. In this previous work, the emphasis was on derivation oinstrument degradation, whereas characterisation of the diffuser BSDF was a necessary byuct of the analysis. For that purpose, it was sufficient to describe the BSDF by a 2nd order pomial function of the azimuthal incidence angle of the solar irradiance. Note that application oBSDF, which constitutes a significant improvement from the on-ground calibration, is curreavailable as option in the GOME Dataprocessor Exctraction software [R4].

In Annex A we have reproduced some relevant figures based on GDAQI work. Annex A sfor each channel the residual solar spectrum as function of time. The residual is defined hthe spectrum normalised to the solar spectrum of 3 July 1995, after correction for BSDF andradation. Especially in channels 3 and 4 a seasonal pattern is clearly recognisable in the residuA close look to channel 1 also reveals a hint of seasonal dependence in the residuals. In chaseasonal dependence in the residuals is not apparent, but here the picture is dominated bin the wave-like patterns. These wave-like patterns are changes in the etalon [R5]. Especichannel 2 these changes were frequent, due to a vacuum leak which caused changing oflayer on the detector after each switchoff of the instrument. In Annex B.5 of the GDAQI resome further figures are presented (not reproduced here), which show for selected wavelenseasonal dependence of the irradiance and the derived BSDF. Also there it is clear that stabsonal patterns exist which are not properly described by the current BSDF calibration. Calib

CHEOPS-GOME: BSDF seasonal effectsCH-TN-DLR-GO-0001Issue 19 May 2004 DLR

4

han-the

DF byref-

ns theegra-find

ave thee peri-rada-

AQI.iduals,dationy aret hasmaxi-

s canndowsepend-aria-apart

bilitytable,ave-it is

errors shown in those pictures are typically up to 0.5% for channels 1 & 2 and up to 1% for cnels 3 & 4, but it is difficult to judge how representative the selected wavelengths are forwhole channel.

1.3 Outline of this study

In the current work, we can achieve further improvement on the seasonal depence of the BSdropping a few restrictions which were set by the scope of the GDAQI project. In GDAQI, theerence date was set to 3 July 1995, but especially in this first phase of GOME operatioinstrument was not as stable as in later years. Also, GDAQI was concerned with long-time ddation, but in later years jumps occurred in instrument degradation, which made it difficult toout the subtle seasonal variations which are present in the data record. In this study, we hfreedom to choose an optimum time window for our analysis, and we can concentrate on thodic effects in the data without the need of providing a parameterisation of instrument degtion.

This study can be divided in two parts. In Section 3 we closely follow the method used in GDHere the main problem in the analysis, which is the changing etalon that dominates the resis removed by smoothing the data over wavelength. This implies that both the derived degraand the derived BSDF are also smooth functions of wavelength. The result from this studsurprising in the sense that the BSDF found is not a smooth function of illumination angle, bua plateau (channels 1 & 2) or even a pronounced dip (channels 3 & 4) at the angle wheremum reflectivity was expected.

In Section 4 we follow a different concept. Here we derive a BSDF for each wavelength. Thionly be done when the changes in etalon are not too pronounced, i.e. for restricted time wiin channels 1,3,4 but not in channel 2. This leads to the surprising result that the seasonal dence of the BSDF is not at all a smooth function of wavelength, but shows large etalon-like vtions with wavelength. Peak-to-peak differences between two wavelengths which are 5 nmcan be as large as 2% in channel 1.

Before we start with the BSDF analysis, we first look at an additional source of error: the staof the Sun. The use of solar irradiance as proxy for the BSDF is only allowed if the Sun is sbut it is known that this is not quite the case below 300 nm. In Section 2 we show that for wlengths below 265 nm the effect is negligible for the years 1995-1998, and for 1999-2002probably negligible for the analysis (monthly variation below the 0.3% level).



5

rradi-angecon-

wave-stance

howsto theormal-onth

thefor

ability

2 Variability of solar irradiance 1995-2002

For this part of the study we used the SOLAR2000 Research Grade V2.23 empirical solar iance model from Space Environment Technologies [R6]. The variability in the 115-400 nm rof this model is based on analysis of UARS SOLSTICE and SUSIM data; the current versiontains analysis up to the year 2002. The model yields spectra at a resolution of 1 nm in thislength range. We use the model in a mode which normalises the intensity to a Sun-Earth diof 1 AU.

A model calculation was made for each day of the year, for the years 1995-2002. Figure 1 smodel results for the wavelength interval 230-400 nm. The data in the figure are normalisedspectrum of January 1st, 1996. One panel is plotted for each year. Each panel shows the nised spectrum at minimum (blue) and maximum (red) intensity level at 230 nm, for six 2-mintervals.

We see significant solar activity effects below 263 nm, near the Mg II line at 280 nm, and inCa II lines near 395 nm. In the remainder of the spectrum, the variability remains below 0.3%the years 1995-1998. For later years, even in these ‘quiet’ regions of the spectrum, the varirises above 0.5%.


6

Figure 1: Model spectra normalised to the spectrum of Jan. 1st, 1996 (intensity ratio versus wavelength).



7

poly-oci-

, solarr may

s are in

timen. Ino mid-talonted byg for.

a 3rdith ased)., butw in

alter-sig-

varia-sub-t.

ve the. Afterility)

ctionthat

ing ofuch

ction

3 Seasonal dependence of the ‘smoothed’ BSDF

3.1 Method

In this Section we present an analysis of the Solar Irradiance after fitting the spectra with anomial in wavelength. This polynomial fit is the only reliable way to remove the problems assated with changes in the Etalon.

Without smoothing, the change in intensity is the product of changes in sun-earth distanceoutput, instrument degradation, change in BSDF, and change in Etalon. The first parametebe calculated. For the timeframe which we have selected (see below), the other parameterthe order of magnitude (see also the GDAQI report):

• solar output: < 0.3% for wavelengths > 263 nm (apart from Mg II and Ca II region)

• instrument degradation: 15% at 260 nm, 5% at 320 nm

• BSDF: up to 1% deviation expected w.r.t. current calibration

• Etalon changes: up to several %

The dominant factor here is the instrument degradation. However, this varies gradually inand can to a large extent be fitted with a polynomial function. This is not possible with Etalochannel 2, one can observe ca. 30 significant changes in etalon structure from mid-1995 t1998 (see Annex A). For a given wavelength, every change yields a jump in intensity. As ecannot reliably be modelled, these jumps cannot be quantified, even if they can be deteceye. These jumps conceal the seasonal variation in BSDF on the level which we are lookinFor this reason Etalon removal is very desirable.

In the data which we show here, we have removed the Etalon by fitting each spectrum withorder polynomial (by ‘spectrum’ we denote here the ratio of a solar irradiance spectrum wreference irradiance spectrum, which is taken as the first spectrum in the time window uAnalysis indicates that using a 4th order polynomial may provide marginally better resultsthis would imply a significant change in the software’s data structure. Furthermore, we shothe next Section that the errors made by any kind of wavelength fitting are much larger. Asnative to polynomial fitting, fourier smoothing was performed but also this did not yield anynificant improvement.

The only change to the GDAQI software made here, was to minimise the influence of solarbility. To this end, the spectral region around the strongly variable Mg II line was cut out andsequently filled-in using a parabolic fit; the region below 250 nm was not taken into accoun

After several trial runs, the period from September 1st, 1995 to May 25, 1998 appeared to gimost stable results. Before this period, outgassing effects in channels 3 and 4 are disturbingthis period, jumps in the irradiance data (jumps in instrument degradation or solar variabwere observed in channel 1 which disturb the periodic effects we are looking for.

It was found that within the variability band, all seasonal changes could be explained as a funof azimuth angle of solar irradiance onto the diffuser, i.e. no periodic signals were founddidn’t correlate with this azimuth angle. This is against the expectation we had at the beginnthe study. The correlation with azimuth angle was found to be surprisingly strong, though mmore complex than expected. For this reason, we will present all results in this study as funof solar azimuth angle.


8

, nor-stance,engthsge valueroundlike(this

e that

well,thenced

etricalnglelowl nearThisectraFigure

s func-lap inngth

a 6thimuthasonal

of thenot bems.tic in

80 nmape as

3.2 Results

Figures 2-5 on the following pages show as function of solar azimuth angle the irradiancemalised to the irradiance of September 1st, 1995, after removal of changes in Sun-Earth diinstrument degradation, and Etalon. The various panels show data for a selection of wavelin each channel. The plotted spectra are rescaled such that for each wavelength, the averaof the spectra in the azimuth interval [-0.1, 0.1] equals 1. The blue line represents the on-gcalibration of the BSDF, and the red line is a second-order polynomial fit which is calculatedthe BSDF correction currently used in the GOME Data Processor Extraction softwaremeans: it is close to, but not necessary fully identical to, the operationally used BSDF sincwas calculated using data from a different time window).

The figures show that in channels 1 and 2 the operationally used BSDF performs quiteexcept for azimuth angles near−4o, where deviations up to 1% can be seen. Surprising is thatBSDF is not a monotonous function of incidence angle on the diffuser, but that a pronoustructure is visible reminiscent to ‘side-lobe’ interference patterns in Fabry-Perot cavities.

The side-lobes have a minimum near 300 nm, and here the BSDF is also the most symmw.r.t. azimuth angle. Note, that as function of wavelength, the curves ‘pivot’ around azimuth a0 (the location of the pivot point may be determined by the normalisation to this angle): forwavelengths the BSDF is lower for the negative azimuth angles, then becomes symmetrica300 nm, and for higher wavelengths the BSDF is lower for the positive azimuth angles.implies that, cycling over azimuth angle, the spectral shape of the GOME irradiance spbecomes alternatingly convex and concave over the course of the year. This is apparent in6.

In channels 3 and 4 the side lobes near azimuth angles of +/−4o are particularly pronounced. Notealso the smaller-scale wobbles on the curves. These features are remarkably reproducible ation of azimuth angle. Even at the beginning of channel 3, where the spectra do not overabsolute intensity (probably due to dichroic shifts which lift the 3rd order polynomial wavelefit), the small-scale structure as function of azimuth angle is very similar for all spectra.

Within the variation of the curves, a good fit can be obtained for channels 1 and 2 usingorder polynomial in azimuth, for each wavelength. It appears that each coefficient for the azfit can be described to high accuracy by a 3rd order polynomial in wavelength. Thus the sedependence of the BSDF in channels 1 and 2 can be expressed by4 x 7 = 28coefficientsCij ; ingeneral we obtain:

where is a ‘central wavelength’ in the channel (to keep the numerical values ofCij in a smallerrange of magnitude); n=6 for channels 1 and 2, and n=8 for channels 3 and 4.

For channels 3 and 4, an 8th-order polynomial fit is needed to describe the sidelobe shapecurves; higher orders do not bring significant improvement. The smaller-scale structure canfitted by a polynomial. It is also too irregular to be fitted by a limited number of fourier terFurthermore it must be noted that due to the dichroic shift, the etalon smoothing is problemachannel 3 up to 480 nm and in channel 4 up to 620 nm; also the end of channel 4 near 7appears to be problematic. The problems here lie in the absolute values of the BSDF; the shfunction of solar azimuth angle can still be fitted fairly accurate.

BSDF λ αazi,( ) Cij λ λc–( ) j⋅j 0=

3

∑

αazii⋅

i 0=

n

∑=

λc



9

Channel: 1 252.60420nm

-6 -4 -2 0 2 4 6Solar Azimuth angle

0.970

0.980

0.990

1.000

rel.

inte

nsity



0.970

0.980

0.990

1.000

rel.

inte

nsity



0.970

0.980

0.990

1.000

rel.

inte

nsity



0.970

0.980

0.990

1.000re

l. in

tens

ity



0.970

0.980

0.990

1.000

rel.

inte

nsity



0.970

0.980

0.990

1.000

rel.

inte

nsity



0.970

0.980

0.990

1.000

rel.

inte

nsity



0.970

0.980

0.990

1.000

rel.

inte

nsity

Figure 2: Normalised intensity at 8 wavelengths in channel 1 as function of solar azimuth angle


10



0.970

0.980

0.990

1.000

rel.

inte

nsity



0.970

0.980

0.990

1.000

rel.

inte

nsity



0.970

0.980

0.990

1.000

rel.

inte

nsity



0.970

0.980

0.990

1.000re

l. in

tens

ity



0.970

0.980

0.990

1.000

rel.

inte

nsity



0.970

0.980

0.990

1.000

rel.

inte

nsity



0.970

0.980

0.990

1.000

rel.

inte

nsity



0.970

0.980

0.990

1.000

rel.

inte

nsity




11



0.970

0.980

0.990

1.000

rel.

inte

nsity



0.970

0.980

0.990

1.000

rel.

inte

nsity



0.970

0.980

0.990

1.000

rel.

inte

nsity



0.970

0.980

0.990

1.000re

l. in

tens

ity



0.970

0.980

0.990

1.000

rel.

inte

nsity



0.970

0.980

0.990

1.000

rel.

inte

nsity



0.970

0.980

0.990

1.000

rel.

inte

nsity



0.970

0.980

0.990

1.000

rel.

inte

nsity



12



0.970

0.980

0.990

1.000

rel.

inte

nsity



0.970

0.980

0.990

1.000

rel.

inte

nsity



0.970

0.980

0.990

1.000

rel.

inte

nsity



0.970

0.980

0.990

1.000

rel.

inte

nsity



0.970

0.980

0.990

1.000

rel.

inte

nsity



0.970

0.980

0.990

1.000

rel.

inte

nsity



0.970

0.980

0.990

1.000

rel.

inte

nsity



0.970

0.980

0.990

1.000

rel.

inte

nsity




13

Relative BSDF for several azimuth angles

240 260 280 300Wavelength [nm]

0.960

0.970

0.980

0.990

1.000

1.010

-6.5 -6.0 -4.0 -2.0 0.0 2.0 4.0 6.0

Figure 6: Spectral shape of the ‘smoothed’ BSDF seasonal dependence in the region 250-305 nm, for 8different solar azimuth angles. The curves have been normalised to the BSDF at azimuth angle 0.


14

possi-

talonpat-time.

ithoutlong-deg-

ths in

r toeasons,time)rsus

evel isen ine cal-teady

995),

se thering

quen-urier

se alll (the

solarDF atd 8th

ime

muthn nor-

s over

4 Seasonal dependence of the BSDF without wavelength smoothing

The wavelength smoothing which has been applied so far to remove Etalon, makes it imble to detect small-scale spectral variations which may be present in the data.

A closer look to the figures of Annex A, shows that for channels 1,3, and 4, the long-term Epattern it quite stable, with a few temporary shifts in Etalon - where after a while the originaltern is restored. On the long term, the Etalon pattern seems to deepen but not to shift with

This opens the possibility to derive the seasonal variation in BSDF for each wavelength, wwavelength smoothing. The long-term deepening of the etalon can then be described by aterm degradation correction which is derived for each wavelength. This results in a long-termradation which will be less for wavelengths at the Etalon maxima, and stronger for wavelengthe minima of the Etalon curve. See Figure 7.

After correction for long-term degradation in this way, we can obtain plots which are similaFigures 2-5 but show a much larger scatter amongst the BSDF values for subsequent swhich is due to the fact that any shifts in Etalon cause jumps in the intensity (as function offor a particular wavelength. Nevertheless, we can still derive a meaningful fit of BSDF vesolar azimuth angle: the shape of this curve remains well-defined, although the absolute lmore uncertain. This is similar to the dichroic shift problem in the beginning of channel 3 seFigure 4. In fact, for the beginning of channel 3, the scattering now becomes smaller, as thculation of long-term degradation without wavelength smoothing can compensate for a sdichroic shift.

The procedure in detail is as follows:

1. Interpolate all solar irradiance spectra to the wavelength grid of the first one (from 1.9.1and divide by the first spectrum.

2. Remove high-frequency noise, which is introduced by the fraunhofer structure becauspectra are not perfectly aligned in wavelength. This is done by low-pass fourier filteusing the same method as the etalon filtering described in [R7] (except that the low frecies are not cut out). The cut-off filter frequencies are chosen from the point where the fotransform of the summed normalised spectra start to show white noise.

The remainder is done for each detector pixel:

3. Derive a first-guess long-term degradation using only the points at azimuth angle 0 (thehave the same seasonal BSDF). This degradation is fitted with a 3rd order polynomiaresult for the pixels in channel 1 is depicted in Figure 7).

4. Divide the data from 2. by the long-term degradation and perform a polynomial fit overazimuth angle, to determine the seasonal variation in BSDF (a normalisation to the BSazimuth angle 0 is taken). A 6th order polynomial was finally used for channels 1,2 anorder for channel 3,4.

5. Divide the data from 2. by the BSDF and perform a polynomial fit of high order to the tsequence of spectra, to determine the long-term degradation.

An iteration is performed over steps 4 and 5; three iterations are found to be sufficient.

The results are shown in the following set of figures. Figs. 8 and 9 show for 8 different aziangles the seasonal variation of the BSDF as function of wavelength. The curves have beemalised to the BSDF at azimuth angle 0. The dotted lines represent 3rd order polynomial fitwavelength for each azimuth angle.



15

thermal-wave-

For completeness, the full picture is provided in the 3-D plots of Figs 10 - 13. This shows inupper panel the seasonal variation in BSDF calculated with the method of this Section (noised at azimuth angle 0), and in the lower panel the same surface after fitting the data in thelength direction by a 3rd order polynomial.

200400

600800

1000

260270

280290

300

0.75

0.80

0.85

0.90

0.95

1.00

1.05

Wavelength

Days since 1.9.1995

Figure 7: Long-term degradation in channel 1 (corrected for seasonal dependence of the BSDF):Third-order polynomial fit in time for each wavelength showing deepening of the Etalon


16

Channel: 1 BSDF versus wavelength

240 260 280 3000.960

0.970

0.980

0.990

1.000

1.010

-6.5 -6.0 -4.0 -2.0 0.0 2.0 4.0 6.0


320 340 360 380 4000.960

0.970

0.980

0.990

1.000

1.010

-6.5 -6.0 -4.0 -2.0 0.0 2.0 4.0 6.0

Figure 8: Seasonal variation in BSDF for channels 1&2 for 8 azimuth angles, normalised to azimuth=0



17


400 450 500 550 6000.960

0.970

0.980

0.990

1.000

1.010

-6.5 -6.0 -4.0 -2.0 0.0 2.0 4.0 6.0


600 650 700 7500.960

0.970

0.980

0.990

1.000

1.010

-6.5 -6.0 -4.0 -2.0 0.0 2.0 4.0 6.0

Figure 9: Seasonal variation in BSDF for channels 3&4 for 8 azimuth angles, normalised to azimuth=0


18

Channel: 1 BSDF unsmoothed

-6. -4.

-2. 0.

2. 4.

6.

260. 270.

280. 290.

300. 0.960

0.970

0.980

0.990

1.000

1.010

Channel: 1 BSDF unsmoothed + polyfit over wv

-6. -4.

-2. 0.

2. 4.

6.

260. 270.

280. 290.

300. 0.960

0.970

0.980

0.990

1.000

1.010

1.020

Figure 10: Seasonal variation in BSDF for channel 1, as function of solar azimuth angle and wavelength.Data have been fitted with 6th order polynomial in azimuth, and normalised to the function at azimuth=0



19


-6. -4.

-2. 0.

2. 4.

6.

320. 340.

360. 380.

0.970

0.980

0.990

1.000

1.010


-6. -4.

-2. 0.

2. 4.

6.

320. 340.

360. 380.

0.975

0.980

0.985

0.990

0.995

1.000

1.005



20


-6. -4.

-2. 0.

2. 4.

6.

450. 500.

550. 0.985

0.990

0.995

1.000

1.005

1.010

1.015


-6. -4.

-2. 0.

2. 4.

6.

450. 500.

550. 0.985

0.990

0.995

1.000

1.005

1.010

1.015




21


-6. -4.

-2. 0.

2. 4.

6.

650. 700.

750. 0.980

0.990

1.000

1.010

1.020


-6. -4.

-2. 0.

2. 4.

6.

650. 700.

750. 0.985

0.990

0.995

1.000

1.005

1.010

1.015



22

r azi-.

atureency

in the

a andphase-etalonor thesponseure onormal-

howignifi-

azi-

mial.

h azi-ottedfrome pol-s ontalon

veri-nels 1l 1 theof theow-r var-thechan-

iably

5 Discussion

5.1 Summary of results

This work shows two surprising effects in the seasonal variation of the GOME BSDF:

• The BSDF varies not as a simple monotonically increasing or decreasing function of solamuth angle, but has wiggles which point to some kind of interference effect taking place

• The BSDF varies not as smooth function of wavelength, but shows a strong etalon fewhich varies with solar azimuth angle. Channels 3 and 4 show in addition high-frequwiggles.

Regarding the azimuth behaviour, the difference with a second order polynomial fit as usedoperational GDP is of the order of +/−0.5%.

For channel 1, the etalon structure may be as large as 2% at azimuths below−6o. Interestingly,Fig. 8 shows that above 280 nm the Etalon, normalised to azimuth angle 0, has its minimmaxima more or less aligned for all angles, at least for channels 1 and 2. There is a someshift for the largest azimuth angles, and a larger phase-shift for incidence angle -2, but theamplitude relative to angle 0 is quite small. In channel 1 above 280 nm, the etalon structure fsmaller incidence angles correlates with the (much stronger) etalon on the Radiance Refunction. Assuming that the absolute (unnormalised) Etalon correlates with the etalon structRadiance Response, the alignment of the etalons for various angles indicates that the unnised Etalon pattern will have a minimum amplitude near azimuth angle 0.

5.2 Reliability of the results

The question is how reliable the method of determining the ‘unsmoothed’ BSDF is, i.e.insensitive it is to changes in Etalon pattern. After all, the changing Etalon patterns cause scant jumps in intensity for each pixel, and the question is here if the periodic changes withmuth are strong enough to be detected despite these jumps.

One test is to see if the method of determining the ‘unsmoothed’ BSDF yields, after polynofitting over wavelength, the same result as the method of determining the ‘smoothed’ BSDF

The dotted lines in Figs 8 and 9 represent 3rd order polynomial fits over wavelength for eacmuth angle. If the fits are not influenced significantly by shifts in etalon structures, these dlines should be identical to the ‘smoothed’ BSDF in Fig. 6. This is indeed the case, apartsome small differences at the fit window edges which may be attributed to uncertainties in thynomial fitting of etalon structures (note that for the ‘smoothed’ BSDF we fit etalon structurethe measurement spectra, while for the wavelength fit of the ‘unsmoothed’ BSDF we fit estructures on the BSDF).

Another test is to do the calculations described in Section 4 in another time window. For thisfication we chose the time interval between June 1996 and June 1999. The result for chanand 2 are shown in Fig. 14 , which has to be compared to Fig. 8. This shows that for channeetalon structure is basically the same. The amplitude of the etalon, and also the amplitudemean seasonal variation, is slightly larger in this verification than in the original calculation. Hever, the scatter in the relative intensities (due to non-uniform instrument degradation or solaiability) is so much larger that the difference is not significant (with relative intensity we meancurves in Figs. 2-5). Significant is that the Etalon pattern is the same. This is not the case fornel 2, which confirms our expectation that for channel 2 only a ‘smoothed’ BSDF may be rel



23


240 260 280 3000.960

0.970

0.980

0.990

1.000

1.010

-6.5 -6.0 -4.0 -2.0 0.0 2.0 4.0 6.0


320 340 360 380 4000.960

0.970

0.980

0.990

1.000

1.010

-6.5 -6.0 -4.0 -2.0 0.0 2.0 4.0 6.0

Figure 14: Results from the non-optimal time window used for verification


24

sicallyities is

large,1.

struc-e irra-n error

tio off theseibration

e withs arennel 1,DF, asthat in

d Solarr byysis isnce).

eas-the fitm deg-ectralak-to-ork asonsist-lso ber work

m the

ntlye dif-F. A

iancess been

riationgth-lon on

calculated because of the frequent etalon shifts. For channels 3 and 4 the verification bayields the same results as the original calculation; also there the scatter in relative intensmuch larger than the difference in derived BSDF.

5.3 Relation to Ozone profile retrieval

We expect that the amplitude of the etalon features on the seasonal variation in BSDF is sothat these features cannot be neglected for Ozone profile retrieval, at least not in Channel

A limitation of the study in this report is that we can derive the seasonal dependence of thetures on the BSDF, but we cannot derive the absolute value of the BSDF. This is because thdiance measurements yield the product of radiance response times BSDF, and the calibratioin the radiance response is not known. The SCIAMACHY calibration has show that the rairradiance calibration to radiance calibration show etalon features, but the exact structure ofeatures was dependent on the exact calibration setup (e.g. changes in distance to the callamp caused changing etalons on the data).

Nevertheless, some indications may be gained from comparisons of the GOME irradiancSOLSTICE irradiance as presented in the GDAQI report. Here the ratio of both irradianceshown for several solar azimuth angles. One can observe changing etalon patterns in chabut those in channel 2 seem to be more stationary. This does not proof anything for the BSSOLSTICE does not measure Earthshine radiance, but it may be taken as an indicationchannel 2 the seasonal variations in etalon structure are not as large as in channel 1.

As there are no other space-born instruments which measure both Earthshine radiance anirradiance at the required wavelength resolution (apart from SCIAMACHY, which is similadesign and is expected to have similar uncertainty), the only remaining absolute BSDF analto compare GOME measurements by modelled reflectivities (reflectivity = radiance / irradia

Van der A [R8] and Latter et al. [R9] have performed Ozone profile retrievals from GOME murements over several years, where they derive correction factors on the reflectivity fromresiduals. Note that these residuals may not only contain seasonal effects, but also long-terradation effects of the BSDF. The average residual found by Van der A contains some spstructure which seem to correlates with the etalon structure found here, although his 3% pepeak error bar does not allow hard conclusions. Latter et al. interpret the residuals in their wfollows: “The quasi-sinusoidal time dependence and phase delay .... appear qualitatively cent with the behaviour of an etalon”. Both authors mention that the etalon structure may adependent on incidence angle on the scan mirror. Such a dependence cannot follow from ouhere, as we always use the scan mirror position of the solar calibration which is different fropositions at the radiance measurements.

Unfortunately, according to [R10], the ozone retrievals of Latter et al. do not improve significawhen they use the BSDF correction factor derived by Van der A, which may indicate that thferent retrieval schemes find different compensations in their retrieval for errors in the BSDmore reliable way to derive an absolute BSDF seems therefore to compare GOME raddirectly to modelled radiances using known, measured, ozone profiles. This approach haproposed by the SRON team in the framework of the CHEOPS project.

Concluding we can say, that this study clearly shows the existence of complex seasonal vain the BSDF, with a significant variability in etalon structure. We believe that the wavelensmoothed seasonal variation in BSDF has been accurately modelled, but the absolute EtaBSDF must be derived by other means.



25

olar-

t

5,

meas-

ast

-

val",

da-

6 Reference Documents

[R1] S. Slijkhuis, "GOME instrument properties affecting the calibration of radiance and pisation", GOME Geophysical Validation Campaign

[R2] A. Richter and T. Wagner, "Diffuser Plate Spectral Structures and their Influence onGOME Slant Columns", Technical Note, Institut für Umweltphysik Univ. Bremen andInstitut für Umweltphysik Univ. Heidelberg, January 2001

[R3] I. Aben, M. Eisinger, E. Hegels, R. Snel, C. Tanzi, "GOME Data Quality ImprovemenGDAQI Final Report", SRON report TN-GDAQI--003SR/2000, 29.9.2000

[R4] GOME Data Processor - Extraction Software User's Manual, ER-SUM-DLR-GO-004Issue 1, 4.8.1999

[R5] Jochen Stutz and Ulrich Platt, "Problems in using diode arrays for open path DOAS urements of atmospheric species", Institut für Umweltphysik, Universität Heidelberg

[R6] W. Kent Tobiska et al., The SOLAR2000 emperical solar irradiance model and forectool, J. of Atm. and Solar-Terrestial Physics 62, p.1233, 2000

[R7] SCIAMACHY Level 0 to 1c Processing - Algorithm Theoretical Basis Document, ENVATB-DLR-SCIA-0041, Issue 1, 19.2.1999

[R8] R. van der A, "Recalibration of GOME spectra for the purpose of ozone profile retrieKNMI Technical Report TR-236, 26.6.2001

[R9] B. Latter, R. Siddans, B. Kerridge, "New Investigation of GOME-1 Scan-Mirror Degration", Draft version, 19.11.2002

[R10] B. Kerridge, private communication, April 2004


26

LR’s[R3]

ction 4

rectedn cor-

hite

ible inindi-d side

7 Annex A: Relevant figures based on GDAQI work

On the following pages, we attach to this report figures reproduced from version 1.07 of DGOME instrument degradation report. These figures are similar to those of the GDAQI studywhich have been used as the starting point for our work. These figures can be found in Seof the GDAQI report on pages 71-74.

The four figures are for channels 1 to 4 respectively.

For each channel is shown:

Ratios of GOME solar spectra with the spectrum of July 3, 1995. These ratios have been corfor long-term degradation and normalised to a Sun-Earth distance of 1 AU; they have beerected for the BSDF which was derived in the GDAQI study.

Intensity is color-coded; the data are as function of wavelength and as function of time. Wlines represent missing or corrupted GOME solar measurements.The wave-like patterns are etalon structures relative to July 3, 1995. Changes in etalon vischannel 2 correlate with detector-cooler switch-offs; the time of switchoffs is not separatelycated here, but can be found in the GDAQI report where they are indicated on the righthanbar of the figures.



27


28



29


30