LUNAR CALIBRATION OF MSG/SEVIRI SOLAR CHANNELS
1
1
1
2
Bartolomeo Viticchiè , Sébastien Ch. Wagner , Tim J. Hewison , Thomas C. Stone ,
1
1
1
1
Jagjeet Nain , Rebeca Gutierrez , Johannes Müller , and Christopher Hanson
1
EUMETSAT, Eumetsat Allee 1, 64295 Darmstadt, Germany
US Geological Survey, 2255 North Gemini Drive, Flagstaff, AZ 86001
2
Abstract
We present the main results of the study performed at EUMETSAT on the stability of MSG/SEVIRI
solar channels using the Lunar Calibration Prototype based on our fully validated implementation of
the USGS ROLO model. The study has been undertaken on an archive of lunar observations from
MSG1 (between 2003 and 2013) and MSG2 (between 2006 and 2013) in all the SEVIRI Solar
Channels and for both Full Disc and Rapid Scan Service modes. The completeness of the
observational conditions in this archive is unprecedented: it is a unique collection of Moon
observations covering a wide range of both phase angle (illumination) and libration angles.
The results of the Lunar Calibration have a phase angle-dependence which introduces extra scattering
in the calibration results. This dependence has been observed and quantified for the very first time
using the Moon observation archive available at EUMETSAT, and has a larger impact at long
wavelengths (around 1.6 µm) than short ones (around 0.6 µm).
By correcting for the phase angle-dependence the performance of SEVIRI can be measured with a
-2
precision of 10 % per year. In its four solar channels SEVIRI is stable within 0.7% per year (i.e., the
largest drift measured for MSG1 and MSG2), fulfilling the stability requirement of 2% per year. This
result is found by i) estimating the irradiance bias with respect to ROLO (which includes instrument
natural drift plus the calibration) and ii) deriving the gain for the instruments taking ROLO as a
reference (this includes only the instrument natural drift).
INTRODUCTION
The SEVIRI (Spinning Enhanced Visible and Infrared Imager) aboard the Meteosat Second
Generation (MSG) platforms scan the Earth from Geostationary orbits producing images in twelve
spectral bands in the wavelength range 0.56-14.4 μm. SEVIRI has four solar bands, namely, the
VIS06, the VIS08, the NIR16, and the High spatial Resolution VISible (HRVIS; Schmetz et al., 2002).
For these bands absolute calibration accuracy of 10% for short-term applications and 2% (per year)
for long-term stability are required.
The calibration of the SEVIRI solar bands is based on a vicarious calibration technique using stable
desert scenes as transfer targets (Govaerts et al., 2004). Such a method provides the calibration
coefficients and allows one to monitor the sensor temporal drift. However, several years are required
to derive reliable drift estimates and to reduce the uncertainties caused by seasonal variations and
changes in surface properties of the targets.
To overcome these limitations the Moon can be employed as a complementary target for drift
monitoring. The Moon is a target whose properties are exceptionally stable in time (Kieffer, 1997), and
since it crosses regularly the SEVIRI field of regard it can be used as a radiometric reference by
means of a lunar irradiance model.
The ROLO model (Kieffer & Stone, 2005) allows one to compute the total irradiance from the Moon
disc for a given position of the observer and observation time. It can reproduce the combined effects
of the lunar phases, the non-Lambertian reflection on the surface of the Moon, and the lunar librations.
An important point to stress is that due to the stability of the lunar surface properties (Kieffer, 1997) the
ROLO model can be used as radiometric reference for observations made at any time.
The ROLO calibration model has been already used for different calibration purposes: calibration of
SeaWiFS (Eplee et al., 2012), cross-calibration of SeaWiFS-MODIS (Eplee et al., 2011), calibration of
Mercury observations from the MESSENGER mission (Holsclaw et al., 2010). Figure 8 of Eplee et al.
(2011) represents the ROLO calibration results for different instruments as a function of the phase
angle in the range ±90 deg (i.e., the validity range of the ROLO model; Kieffer & Stone, 2005). This
shows that the difference between the observed irradiance and ROLO irradiance seems to vary with
the phase angle. However, because of the large scatter of the calibration results of each dataset there
was no firm conclusion on a possible phase angle-dependence of the bias. Here we show for the first
time that there is a clear phase angle-dependence of the results of the ROLO calibration and that this
dependence can be quantified and handled properly to derive a precise measure of the stability of the
solar channels of SEVIRI.
LUNAR OBSERVATION ARCHIVE
The archive of lunar observations was created (and will continue to be populated in the future) by
following a three-step procedure for each operational satellite:
1. Look for the acquisitions in which the Moon is in the SEVIRI field of regard.
2. Extract the portion of data-image including the Moon (hereafter the Moon imagette) from the
Level 1.0 data (in Digital Counts), and convert this to Level 1.5 radiance.
3. Gather the extracted Level 1.0 Digital Counts imagette, the Level 1.5 radiance imagette, and
all the information needed by the Lunar Calibration Prototype to perform the calibration.
The data archive is currently composed of 542 low-res (Moon available in all low-res channels) plus 70
HRVIS imagettes from MSG1 (covering the time window 2003 – present), 420 low-res plus 91 HRVIS
imagettes from MSG2 (covering the time window 2006 - present), and few low-res imagettes from
MSG3 (here we do not present the analysis of these imagettes). For both MSG1 and MSG2 the
number of lunar observations per year is about 60. This dataset is a unique collection of Moon
observations covering a wide range of both illumination and libration angles. In Figure 1 and Figure 2 we
represent the simultaneous observation of the Moon in the three low-res channels and the high-res
channel, respectively.
Figure 1: Example of Moon radiance imagettes in the three low-res channels VIS06, VIS08, and NIR16 (from left to
right). Each imagette has been scaled between its min and max.
Figure 2: Example of Moon radiance imagette in the HRVIS channel scaled between its min and max.
The smaller number of occurrences for the HRVIS channel compared to the low-res ones can be
explained by pointing out that the high-resolution channel observes only a portion of the total field of
regard of SEVIRI.
LUNAR CALIBRATION PROTOTYPE
shows a flow chart of the Lunar Calibration Prototype based on the ROLO model implemented
at EUMETSAT in collaboration with USGS and tested against the USGS Lunar Calibration. The ROLO
model provides 1% relative accuracy and is consequently used for monitoring the temporal stability of
SEVIRI.
Figure 3
Figure 3: Flow chart of the Lunar Calibration Prototype implemented at EUMETSAT.
The lunar calibration starts with the processing of the data (the blue DATA block in the chart); this
processing takes place over three branches:
1. The Integration over the Moon imagette in Radiance: this step allows one to derive the total
irradiance, i.e., IrrOBS.
2. The extraction of the MSG position at the time of the Moon observation: the position of the
satellite is expressed in the Earth-fixed reference ITRF93.
3. The Evaluation of the deep space level in Digital Counts (DC): this is the average of the
counts from the deep space pixels taken from each imagette and provides the offset in DC
(OffDC).
The quantities derived from the blue branch number 2 are used as inputs for the yellow
OBSERVATION
GEOMETRY
block
which
is
built
around
the
SPICE
Toolkit
(http://naif.jpl.nasa.gov/naif/toolkit.html). This block allows one to derive two groups of quantities
needed to use the ROLO model:
1. the selenographic longitude and latitude of the Sun and MSG at the time of the observation of
the Moon (these are coordinates in the Moon-Mean Earth Rotation Axis reference
MOON_ME), and
2. the distances Sun-Moon and Moon-MSG at the time of the observation.
The selenographic coordinates from the yellow branch number 1 are used to derive the disc-integrated
Moon spectral reflectance by using the ROLO model (Ak, see Eq. 10 in Kieffer & Stone, 2005). This,
together with the Solar Spectral Irradiance (SIrr) and the Spectral Response Function of each channel
(SRF) are used to derive the ROLO reference irradiance at standard distances (1 AU and 384400 km):
-2
Eq. 1
-1
[mW m µm ],
where λi represents a generic wavelength among the Nλ wavelengths of the
Spectral Response Function. We specified here the way the irradiance at
standard distances is derived since it differs from the method described in
Kieffer & Stone (2005).
By using the distances from the yellow branch number 2 one can perform the re-scaling of the ROLO
irradiance at real distances to derive the ROLO reference irradiance (IrrROLO).
-2
-1
[mW m µm ],
Eq. 2
where
is the distance between the satellite (MSG) and the Moon in km and
is the distance between the Sun and the Moon in AU.
In the next section we present the two main outputs of the calibration procedure which are derived by
employing
,
and
.
Computation of the calibration results
By using the IrrOBS and the IrrROLO one can derive:
Eq. 3
[%].
ΔIrr is a measure (in percentage) of the bias between observations and the ROLO model, which is a
measure of the performance of the combination instrument-calibration.
By using
and
one can derive the Slope (gain) which would give the ROLO irradiance
(i.e., Slope at the ROLO scale):
-2
-1
-1
-1
[mW m µm DC sr ],
Eq. 4
where ΩPIX is the SEVIRI pixel solid angle, and NPIX is the number of pixels in
the imagette.
The Slope in Eq. 4 can be used to monitor the instrument drift.
The computation of the quantities in Eq. 3 and Eq. 4 is performed for each Moon observation which
satisfies the following quality criteria:
 The extraction must be completed successfully in all the low-res channels.
 The conversion from DC to radiances must not introduce any Infinite or NaN.
 The complete Moon disc must be in the imagette. The ROLO model provides the discintegrated reflectance, from this the need of having the complete Moon disc observed in the
imagette.
 The Moon must be the only object available in the imagette. The Moon extraction procedure
developed at EUMETSAT allows one to store images in which both the Moon and (a tiny
portion of) the Earth are observed. However, a method to handle the content of these cases
has not been developed yet.
RESULTS
The measure of the stability of the SEVIRI solar channels through the Lunar Calibration Prototype is a
two step process.
1. The phase angle dependence of the results must be measured separately for each channel,
2. the results (i.e., the ΔIrr and the Slope) corrected for the observed phase angle dependence
can be derived
In the next two sub-sections these two steps are described using the MSG2/NIR16 as an example.
Phase-angle dependence: the MSG2/NIR16 case
The phase angle dependence of the ROLO calibration can be visualized by representing ΔIrr vs. g
(see Eq. 3, where g is the phase angle expressed in degrees) as done in Figure 4. This specific case
chosen for the representation is the MSG2/NIR16 channel case, the one for which the phase angle
-1
dependence of ΔIrr is the largest: 6.1 % (90 deg) . Such dependence is measured by performing the
linear fit ΔIrr(g) = a |g| + b, i.e., symmetric with respect to the zero (where a is the variation of ΔIrr over
1 deg due to the phase angle dependence).
Figure 4: ΔIrr vs. g for the MSG2/NIR16 channel (red circles). Both the linear fit (dashed line) and the value of the slope
derived from the fit itself are reported in the plot.
In the following section the impact of the phase angle dependence on the results of the calibration for
the MSG2/NIR16 ΔIrr case is shown and a method to correct for the effect of the dependence is put
forward. This is the method which is currently adopted in EUMETSAT in the calibration of MSGs to
handle the phase dependence in both the ΔIrr and the Slope measure.
ΔIrr corrected for the phase angle dependence: the MSG2/NIR16 case
The ΔIrr as a function of the Moon observation time for the MSG2/NIR16 channel is represented in
both Figure 5 and Figure 6. Figure 5 shows the result before applying the correction for the phase
angle dependence. A systematic change of the phase angle from small (dark red) to large values (light
orange) is found from the bottom to the upper part of the plot. Besides this, the MSG2/NIR16 is found
-1
to be very stable in time: 0.045 ± 0.066 % year .
Figure 5: ΔIrr as a function of the Moon observation time expressed in Years (coloured dots) for the MSG2/NIR16
channel. The colour code represents the absolute value of the phase angle in the validity range of the ROLO model:
between 2 deg and 92 deg (absolute value). The measure of the scatter of the points with respect to the average value
(σ) together with the result of the linear fit performed to measure the annual drift (solid line) are reported.
Figure 6 shows the ΔIrr after correcting for the phase angle dependence by using the slope from the
linear fit in the previous sub-section. The impact of the correction on the result is large and can be
easily quantified by comparing the two values of the residuals’ standard deviation, σ, before and after
applying the correction: the scatter is reduced from 2.02% to 0.89%. Besides, the measure of the
annual drift can be realized with a slightly higher precision after the correction. The MSG2/NIR16 is
found to be very stable also after the correction for the phase angle dependence: 0.040 ± 0.029 %
-1
year .
Figure 6: ΔIrr corrected for the phase angle dependence as a function of the Moon observation time expressed in
Years (coloured dots) for the MSG2/NIR16 channel. Same format of Figure 5.
The results presented in this section can be generalized to all the channels of both MSG1 and MSG2;
the temporal stability of the irradiance measurements fulfils the requirements specified in the
introduction.
Slope corrected for the phase angle dependence: the MSG1/VIS06 case
In Figure 7 the Slope (see Eq. 4) is represented as a function of the Moon observation time. This has
been corrected for the phase angle dependence adopting the same approach described for the ΔIrr.
As pointed out in the section describing the computation of the calibration results, the temporal
variation of the Slope is the measure of the instrument drift (no calibration is involved). This is a drift
evaluation alternative to the one based on the vicarious calibration technique, which uses stable
desert scenes, that provides the official calibration coefficients of the SEVIRI solar channels (see the
introduction section). In Figure 7 the outcomes of these two methods are compared.
Figure 7: Slope variation with respect to the value at the beginning of the time sequence (expressed in %) as a function
of the Moon observation time expressed in Years (blue circles) for the MSG1/VIS06 channel. The result of the linear fit
performed to measure the annual drift is reported. The black dots represent the variation of the calibration coefficient
available in the SEVIRI Level 1.5 Header with respect to the value at the beginning of the time sequence.
The annual instrument drift of MSG1/VIS06 measured using the lunar calibration is 0.481 ± 0.018 %
-1
year , fully in line with the requirements on the instrument performances. Besides, the variation of the
Slope offers one an easy check of the goodness of the drift measurement via vicarious calibration: this
is the one used to derive the calibration coefficients available in the Level 1.5 Header represented in
the plot for a direct comparison. The variation of the calibration coefficients is a step function
representing the up-grade in time of the coefficients to balance the instrument drift measured with the
vicarious calibration: this follows the smooth variation of the Slope derived with the ROLO model.
As stated for the ΔIrr, the results presented in this section can be generalized to all the channels of
both MSG1 and MSG2; the temporal stability of the SEVIRI solar channels fulfils the requirements
specified in the introduction.
DISCUSSION AND CONCLUSIONS
The Lunar Calibration Prototype based on the ROLO model allows one to perform accurate monitoring
of both calibration and instrument performance for the SEVIRI solar channels and to validate their
performance: all the SEVIRI solar channels aboard the platforms MSG1 and MSG2 perform according
to their requirements.
A phase angle dependence of the ROLO calibration has been clearly observed for the first time by
analyzing the unique SEVIRI data. This has subsequently been confirmed by the calibration of
dedicated lunar observations performed with PLEIADES satellites (CNES) in the visible. The effect of
such dependence can be quantified precisely, and it has been shown that it can strongly impact the
calibration of those instruments that, as SEVIRI, cannot choose to observe the Moon at a specific
phase angle (i.e., at a specific illumination). From this stems the correction for the phase angle
dependence to be twofold:
 it allows one to correct for an extra bias in the calibration results, and
 It allows one to reduce the uncertainty on the drift measure when using a set of observations
with phase angle (in absolute value) between 2 deg and 92 deg.
The archive of Moon observations currently available at EUMETSAT keeps growing as newer
observations are acquired. It has been extended to MSG3 for which the results of the lunar intercalibration will be available in a near future. The present archive is a unique dataset which can be
used for multiple purposes, e.g., testing of lunar reflectance model, post-launch evaluation of
Modulation Transfer Functions.
The lunar calibration is a powerful tool that can be used complementarily to other calibration tools for
stability monitoring purposes as shown in Figure 7. This tool will be part of the EUMETSAT monitoring
system for SEVIRI warm channels combining various methods, will allow cross-checks between
methods, and will provide robust drift estimates.
In the future the Lunar Calibration Prototype will be applied to Meteosat Third Generation (MTG) and
possibly to instruments on the EUMETSAT Polar System Second Generation (EPS-SG). A possible
future application could be also the calibration of Meteosat First Generation data for Climate Data
Records.
The phase angle dependence observed with the ROLO model will also be tackled in the near future.
USGS is planning a correction of the model to eliminate such a bias. This means that the archive of
Moon observations available at EUMETSAT could be used to test the effectiveness of the corrections
made to the model.
REFERENCES
Eplee, R. E., Sun, J., Meister, G., et al., 2011, Cross calibration of SeaWiFS and MODIS using onorbit observations of the Moon, Applied Optics, vol. 50, issue 2, p. 120.
Eplee, R. E., Meister, G., Patt, F. S., et al., 2012, On-orbit calibration of SeaWiFS, Applied Optics, vol.
51, issue 36, p. 8702.
Govaerts, Y. M., Clerici, M., and Clerbaux, N., 2004, Operational Calibration of the Meteosat
Radiometer VIS Band, IEEE Transactions on Geoscience and Remote Sensing, vol. 42 issue 9, p.
1900.
Holsclaw, G. M., McClintock, W. E., and Domingue, D. L., 2010, A comparison of the ultraviolet to
near-infrared spectral properties of Mercury and the Moon as observed by MESSENGER, Icarus, vol.
209, issue 1, p. 179.
Kieffer, H. H., 1997, Photometric Stability of the Lunar Surface, Icarus, vol. 130, issue 2, p. 323.
Kieffer, H. H., and Stone, T. C., 2005, The Spectral Irradiance of the Moon, The Astronomical Journal,
vol. 129, issue 6, p. 2887.
Schmetz, J., Pili, P., Tjemkes, S., et al., 2002, An Introduction to Meteosat Second Generation (MSG),
Bulletin of the American Meteorological Society, vol. 83, issue 7, p. 977