Assessment of GOCI Radiometric Products using MERIS, MODIS ...omtab.obs-vlfr.fr/fichiers_PDF/Lamquin_et_al_OSJ_`12.pdf · blue-green wavelengths (Antoine and Morel 1999). Water-leaving
This document is posted to help you gain knowledge. Please leave a comment to let me know what you think about it! Share it to your friends and learn new things together.
Transcript
Ocean Sci. J. (2012) 47(3):00-00http://dx.doi.org/10.1007/s12601-
Available online at www.springerlink.com
Assessment of GOCI Radiometric Products using MERIS, MODIS and Field Measurements
Nicolas Lamquin1*, Constant Mazeran
1, David Doxaran
2, Joo-Hyung Ryu
3, and Young-Je Park
3
1ACRI-ST, 260 route du Pin Montard, 06904 Sophia Antipolis, France2Laboratoire d’Océanographie de Villefranche, Université Pierre et Marie Curie, CNRS, UMR 7093, 06230 Villefranche-sur-Mer,
France3Korea Ocean Satellite Center, KIOST, Ansan P.O. Box 29, Seoul 425-600, Korea
Assessment of GOCI Radiometric Products using MERIS, MODIS and Field Measurements 7
Based on the low to negligible Rrs values measured in the
near-infrared part of the spectrum (λ > 750 nm, see Fig. 2),
the field Rrs spectra proved to be only representative of the
clear to moderately turbid water masses identified on
satellite data, i.e. not of the most turbid waters found in the
East China Sea.
Using the geometry of observation, Rrs values are normalized
into nRrs, to be compared to remote-sensing data.
3. Results
We present comparisons of normalized remote sensing
reflectances (first) and TOA Rayleigh corrected reflectances
(second) between GOCI and the concomitant MERIS and
MODIS data summarized in Table 2. The analysis shows
global maps of reflectances on October 4th 2011 at selected
wavelengths over the GOCI area. Then, more quantitative
views of the discrepancies between sensors are given over
the two W-E and N-S transects. These transect comparisons
are made on several other dates of acquisition to show
behaviors eventually departing from or confirming what is
observed on October 4th.
For the normalized remote-sensing reflectances the influence
of the temporal variability of the GOCI reflectances as well
as the variability induced by the vicarious adjustments of
MERIS and the use of the SWIR bands in the MODIS
retrievals is discussed. Last, gathering all concomitant data
from all selected dates of acquisition into scatter plots
provides a complete quantitative comparison.
Inter-comparisons of the normalized remote-sensing
reflectances
Comparisons between nRrs products from MERIS and
GOCI (Fig. 3) then MODIS and GOCI (Fig. 4) on October
4th 2011 are presented as maps at selected wavelengths: 412,
490, 555 and 660 nm or equivalent. Order of appearance in
Fig. 3 and Fig. 4 reflects the chronological acquisition time
(from left to right).
As time passes by, we notice for example that the horizontal
cloudy sheet (partly composed of relatively thin clouds as
seen in Fig. 1) at about 39° latitude between Korea and
Japan moved eastwards, which is clear from the two GOCI
snapshots and the MODIS snapshot. On contrary MERIS
marine reflectance could be partially retrieved in that area,
although its acquisition is close in time with GOCI first
image (1:49 and 2:16 UTC respectively). This is most
probably because the MERIS Level 2 processing does not
properly detect these optically thin clouds (yet it has been
improved from 2nd to 3rd reprocessing, see Lerebourg and
Bruniquel 2011), as can be seen on the overestimated nRrs
at 665 nm, but as well because the cloud coverage is
extending quickly with time along the day (checked with a
MODIS/Terra image acquired at 1:40 UTC, not shown here).
Also, a flagging of data because of high glint induces blank
areas in the MERIS retrievals on the Eastern part of the
swath, that does not appear in the GOCI and MODIS
retrievals whose geometries of observation do not trigger
high glint conditions in this series of acquisitions.
A first striking observation is the absence of GOCI data,
and in a lesser extent of MODIS data, in the highly turbid
area of the Yangtze delta, while MERIS provides successful
retrievals very near to the shore. The difference between the
MERIS and GOCI acquisition times is too small to explain
this by a change in the cloud amount (what is confirmed by
a visual inspection of the corresponding composite L1
images). It seems that current GOCI atmospheric correction
masks out highly turbid pixels. Comparatively, the less
turbid zone in the West of the South Korean coast, free of
clouds, shows that both GOCI and MERIS retrievals are
successful except in a tiny area witnessing a high load of
sediments where GOCI shows no data again. It is
recommended that the cloud detection threshold be higher
than current threshold to process pixels over turbid waters.
Note also the systematic retrieval of maximum water
reflectance values north of the Yangtze River mouth in the
Yellow Sea. These high values result from the permanent
Fig. 2. Rrs in situ (C-OPS, 18 stations) from the oceanographicfield campaign
8 Lamquin, N. et al.
presence of a well-known maximum turbidity zone
(Beardsley et al. 1985). The origin of suspended particles in
this shallow coastal zone (5-10 m depth on average) is
twofold: (i) direct export of suspended particles from the
Yangtze River and (ii) accumulation and resuspension of
sediments from the Yellow and East China Seas due to the
regional circulation of water masses (main reason explaining
the persistence of the maximum turbidity zone).
Another feature detected on both MERIS and MODIS
nRrs(660) products is a thick ribbon of high values along
the coasts of Korea and Russia, much thicker than in the
GOCI retrievals. This is very likely due to adjacency effects
(reflection by the water body of light coming from the coast
and into the direction of the sensor, as well as altered diffuse
scattering from surroundings), triggered by the forward
scattering geometry on the half-East part of the MERIS and
MODIS swath. On the contrary, geostationary geometry
is closer to the backscattering domain and makes such
contamination negligible. It is even more interesting to
notice that the small adjacency effect observed for GOCI at
2:16 UTC (Fig. 3) decreases at 4:16 UTC (Fig. 4), when the
scattering angle is closer to 180°.
In term of reflectance level, MERIS nRrs values are
systematically lower than GOCI and MODIS ones. MODIS
nRrs values are higher than GOCI values at 412 nm then
progressively become lower as the wavelength goes towards
red. Again, the MODIS 412 nm should not be taken with too
much consideration because of changes in the radiometric
calibration.
A more quantitative analysis is made by overplotting the
Fig. 3. Maps of nRrs (412, 490, 555 and 660 nm from top to bottom) retrieved from MERIS at 01:49 (left) and GOCI at 2:16 (right) onOctober 4th 2011. The sttaight black lines represent two specific transects (one almost N-S and one almost W-E) along which thedifferent satellite products were compared
Assessment of GOCI Radiometric Products using MERIS, MODIS and Field Measurements 9
nRrs values retrieved from GOCI, MERIS and MODIS
data along the two transects (Fig. 5 to Fig. 8) with large
variations depending on water type. Transects are ideal to also
show nRrs from MERIS retrievals not using the vicarious
adjustment gains as well as from MODIS retrievals using
the SWIR bands. In addition, the three GOCI daily products
(2:16, 3:16, 4:16) are shown along with the two MERIS and
the two MODIS products, which finally provide full
quantitative views of all possible discrepancies at separate
wavelengths and for few selected dates of acquisition
providing the most various and complete examples. Some
sensors may not provide data because of failure, cloudiness,
or flagging where others do and some data may seem to be
missing because of a lack of overlapping of the FOVs. On
all figures the vertical scale is similar for each transect but
the geographical extent along the transects is adapted to fill
each figure horizontally.
At 412 nm, MERIS nRrs values are lower than all the
others, but we notice that deactivating the vicarious calibration
brings them much closer to GOCI retrieval (e.g. October 4th
and September 23rd cases). Contrary to what is seen on Fig.
4 from the October 4th case, MODIS nRrs values are not
necessarily higher than GOCI nRrs values at 412 nm as
seen now on top of the transect figures. However, this is
where the maximum differences are observed (up to a factor
of four) between MODIS and GOCI. Overall, the 412 nm
MODIS reflectances show a very divergent behavior along
the W-E transect, i.e. across the sensor swath. Results at 443
nm are qualitatively comparable to those at 412 nm (not
shown). This confirms the impact of the changes in the
radiometric calibration mentioned in Meister et al. 2012 at
412 nm and, to a lesser extent, at 443 nm.
At 490, 555 and 660 nm, an interesting feature is the good
agreement observed between MODIS and GOCI nRrs
Fig. 3. Continued
10 Lamquin, N. et al.
products, with very nice consistency in the spatial dynamics
and the amplitude of the signal (with however sometimes
slightly higher values for GOCI), except for the most turbid
section of the transects. On the other hand, over those turbid
waters, GOCI nRrs, when successful, get much closer to
MERIS retrieval (see e.g. September 23rd on the W-E transect
and October 4th 2011 on the N-S transect), what shows the
sensibility of the retrieval with the different bright water
atmospheric corrections. At 660 nm the discrepancy (Fig.
4) is logarithmic and accentuates the rather small absolute
difference. The use of SWIR bands for the atmospheric
correction is noisier and does not necessarily lead to
improvements. This retrieval option, along with the removing
of the vicarious gains for MERIS, is not considered further
in the present study.
Overall, where the GOCI retrieval does not fail and at
wavelengths for which MODIS nRrs can be considered
valid, the GOCI nRrs reside within the MERIS and MODIS
uncertainties and their spatial evolution is qualitatively well
correlated to those of the two other sensors. This gives credence
to the quality of the GOCI nRrs products compared to other
sensors.
Inter-comparisons using all concomitant data
First results have been presented only for four dates of
acquisition providing the best overlapping of the FOVs for
analyses over maps and transects. However, the other dates
listed in Table 2 also provide concomitant observations
worth of interest. A grouping of all data together allows the
make-up of general scatter plots of nRrs values obtained
from the three sensors and the computation of linear
regressions. On these scatter plots the nRrs values from
Fig. 4. Maps of nRrs (412, 490, 555 and 660 nm from top to bottom) for GOCI at 4:16 (left) and MODIS at 5:00 (right) on October 4th
2011
Assessment of GOCI Radiometric Products using MERIS, MODIS and Field Measurements 11
GOCI (2:16) and MERIS, then GOCI (4:16) and MODIS
are compared. Another specific interest is to compare the
GOCI (2:16) and GOCI (4:16) nRrs values which are
representative of the natural variability of the seawater
reflectance within two hours.
Fig. 9 shows the resulting scatter plots of nRrs at 412,
490, 555 and 660 nm (or equivalent). The color scale of the
density (i.e. the normalized probability of a (x,y) couple to
“drop” in a cell, normalization is made with respect to the
total amount of couples) is logarithmic but the linear regression
is computed from all data without transformation.
These results confirm the preliminary analysis: (i) on
average MERIS nRrs are lower than GOCI, but MERIS 412
nm band shows slightly higher nRrs at higher turbidity; (ii)
MODIS nRrs values show drastic differences at 412 nm
which should not be taken too much into consideration. The
best correlations (r2 higher than 0.9) are obtained at the
longest wavelengths (red part of the spectrum), which are
closer to the NIR domain where aerosols are detected. The
dispersion within the GOCI data (with two hours difference) is
comparable to the dispersion obtained when comparing
GOCI to MERIS and MODIS data. However, it does not
explain the biases observed between sensors since there is
no bias in the GOCI dispersion.
Inter-comparisons of Rayleigh-corrected TOA reflectances
We remind here that in absence of Sun specular reflection
(Sun glint), the Rayleigh corrected reflectance is composed
of different contributors: aerosols as well as multiple scattering
effects between aerosols and Rayleigh, foam and water-
leaving reflectance propagated at TOA level. Because these
contributors all depend on the viewing geometry, it is
difficult to rigorously compare the TOA signal from low
Earth orbit and geostationary sensors. Our analysis is mainly
Fig. 4. Continued
12 Lamquin, N. et al.
Fig. 5. Overplot of nRrs products from GOCI (at 2:16, 3:16, 4:16 UTC, black to grey), MERIS (red; pink w/o vicarious calibration) andMODIS (dark blue; clear blue with SWIR correction), along the West-East transect, at 412 nm (left) and 490 nm (right) on fourcase studies
Assessment of GOCI Radiometric Products using MERIS, MODIS and Field Measurements 13
Fig. 6. Overplot of nRrs products from GOCI (at 2:16, 3:16, 4:16 UTC, black to grey), MERIS (red; pink w/o vicarious calibration) andMODIS (dark blue; clear blue with SWIR correction), along the West-East transect, at 555 nm (left) and 660 nm (right) on fourcase studies
14 Lamquin, N. et al.
Fig. 7. Overplot of nRrs products from GOCI (at 2:16, 3:16, 4:16 UTC, black to grey), MERIS (red; pink w/o vicarious calibration) andMODIS (dark blue; clear blue with SWIR correction), along the North-South transect, at 412 nm (left) and 490 nm (right) onfour case studies
Assessment of GOCI Radiometric Products using MERIS, MODIS and Field Measurements 15
Fig. 8. Overplot of nRrs products from GOCI (at 2:16, 3:16, 4:16 UTC, black to grey), MERIS (red; pink w/o vicarious calibration) andMODIS (dark blue; clear blue with SWIR correction), along the North-South transect, at 555 nm (left) and 660 nm (right) on fourcase studies
16 Lamquin, N. et al.
Fig. 9. Density scatter plot (logarithmic color scale) and corresponding linear regressions between the GOCI (2:16) and MERIS (left),GOCI (4:16) and MODIS (middle), and GOCI (2:16) and GOCI (4:16) (right) nRrs values at 412, 490, 555 and 660 nm (fromtop to bottom)
Assessment of GOCI Radiometric Products using MERIS, MODIS and Field Measurements 17
aimed at checking GOCI TOA signal consistency for
exploitation over turbid waters.
Fig. 10 shows a panel of the ρRC fields on October 4th 2011
for MERIS 1:49, GOCI 3:16 and MODIS 5:00 at 412 and
660 nm. Other wavelengths produce similar patterns (not
shown). The sole use of the GOCI 3:16 data is justified a
Fig. 10. Maps of Rayleigh-corrected reflectances at 412 (left) and 660 nm (right) for MERIS (top), GOCI 3:16 (middle) and MODIS(bottom) on October 4th 2011
18 Lamquin, N. et al.
posteriori by the much smaller temporal variability of the
ρRC fields (see transects analysis below).
Note that a threshold (> 0.2) has been applied to GOCI
ρRC values in order to discard data over land and most of the
clouds (while MERIS and MODIS processors provides ρRC
only over water); this threshold was set as a compromise
Fig. 11. Overplot of Rayleigh-corrected reflectances from GOCI (at 2:16, 3:16, 4:16 UTC, black to grey), MERIS (red) and MODIS(blue) along the West-East transect at 412 nm (left), 490 nm (middle) and 660 nm (right) on four dates (top to bottom)
Assessment of GOCI Radiometric Products using MERIS, MODIS and Field Measurements 19
between high values over water pixels and small values over
cloudy pixels. At each wavelength, ρRC values compare
qualitatively well with however slightly smaller values for
MERIS except at some high value spots apparently
contaminated by clouds.
Again, the two transects provide more quantitative views
Fig. 12. Overplot of Rayleigh-corrected reflectances from GOCI (at 2:16, 3:16, 4:16 UTC, black to grey), MERIS (red) and MODIS(blue) along the North-South transect at 412 nm (left), 490 nm (middle) and 660 nm (right) on four dates (top to bottom)
20 Lamquin, N. et al.
of these discrepancies for all the acquisition dates. Fig. 11
and Fig. 12 show the three GOCI (2:16, 3:16 and 4:16) along
with the MERIS and MODIS ρRC for the same selection of
dates.
The dispersion between the daily GOCI data (2:16, 3:16
and 4:16 UTC) is much smaller than the difference with the
other sensors, except where strong spikes reveal pixels not
corrected for clouds having an effect below the threshold of
0.2. It means that the natural variability of the signal with
time does not explain the discrepancies observed between
GOCI and MERIS/MODIS. Note that spikes on MERIS
data along the transects are also related to thin clouds not
detected in the L2 processing and should not draw our
attention in the present study (this was checked on L1
data).
As previously mentioned, the across track calibration
problem change of MODIS at 412 nm has been identified
on the West-East transect (in particular on September 23rd
and October 4th 2011); hence these precise data should be
considered with caution. However, on the West part of the
transect, and on the whole North-South transect, there is a
good agreement between GOCI and MODIS at 412 nm,
except on April 11th 2011. This is however not true for
MERIS Rayleigh-corrected reflectances, which are lower
(by a factor of 1.5); this is an unexpected difference, which
cannot be explained by the MERIS vicarious adjustment,
not applied at this stage of the processing, and should be
analysed further. The difference is likely due to viewing
angle difference between sensors, which gives different
path lengths and scattering angles. In the swath center of
MERIS or MODIS, GOCI ρRC would give higher values
due to slant view by GOCI. To give order of magnitude, we
have checked with MERIS radiative transfer Look-up
tables that the aerosol reflectance may vary up to +0.01
between view zenith angle of 10° and 45° in forward
geometry (i.e. in the half-west part of MERIS swath) for
aerosol optical thickness up to 0.15 at 865 nm. It is also
worth noting that GOCI ρRC contains the sea-surface
reflection term, which is not the case for MODIS and
MERIS. This would explain qualitatively the differences in
ρRC between sensors.
At 490 nm, a better agreement is found between MERIS
and MODIS and some discrepancies appear with GOCI
(e.g. October 4th) for the clearest waters, while the signal on
turbid waters (most western part of the W-E transect) is
generally in good accordance for all sensors.
At 660 nm, the variations detected by the GOCI, MERIS
and MODIS sensors are quite consistent, with only significant
differences observed for the lowest values (i.e. clearest
waters).
Considering uncertainties in the sensor comparison, due
to GOCI longer path and GOCI sea surface reflection
effects (mainly sky glint), these preliminary results prove
there is no obvious bias in GOCI Rayleigh-corrected
reflectance (hence in TOA radiance) so that this signal can
be used with confidence over turbid waters where current
atmospheric correction fails to retrieve the seawater
reflectance.
Fig. 13. Match-ups between in situ (green), GOCI (black andgrey), MERIS (red) and MODIS (blue) nRrs spectra onSeptember 23rd 2011 (10:30 am, top, and 1:30 pm,bottom). For the satellite spectra: mean in plain lines andmean +/- standard deviation in dashed lines
Assessment of GOCI Radiometric Products using MERIS, MODIS and Field Measurements 21
Quality assessment against field reflectance measurements
Few in situ nRrs measurements carried out during the
2011 oceanographic campaign were concomitant with clear
sky conditions for match-ups with GOCI, MERIS and
MODIS data. If restricting the collocation with satellite data
within a circle of 0.05° latitude/longitude centered on the in
situ observations, only September 23rd provides data from
all sensors.
Fig. 13 shows all of the in situ, GOCI, MERIS and
MODIS spectra obtained for the two in situ acquisitions in
the East China Sea (32°06.459 N, 125°12.150 E, 10:30 am
local time and 32°06.148 N, 125°11.363 E, 1:30 pm local time).
Satellite-derived nRrs spectra are shown as mean as well
as mean +/- standard deviation (dashed lines) of the nRrs
values collected within the circle of 0.05° surrounding the
field measurement.
Due to the time difference (3 hours) imposed for match-
ups between satellite and field data, only measurements
carried out on September 23rd are considered here. The two
field acquisitions on this date were spatially close and logically
the corresponding satellite-derived spectra are mostly the
same. However, the two in situ acquisitions show a decrease
from the 10:30 am shot (top) to the 1:30 pm shot (bottom).
One of those is between the spectra of GOCI and MERIS
(10:30 am shot) and the other one is generally closer to the
MERIS spectrum.
Both GOCI and MERIS show a bulge around 500-600
nm, which can be related to a general bulge in the in situ
spectra. This bulge is not seen on MODIS since the 412 and
443 nm nRrs are overestimated.
These spectra, only observed on one day of acquisition,
recall some of the features observed along the previous
analyses of nRrs: at short visible wavelengths (412-443
nm), MODIS nRrs values are overestimated while MERIS
nRrs seem sometimes too low.
In this particular case the in situ measurements give
highest confidence in the MERIS retrieval (spectrum 2 of
Fig. 13) and, to a lesser extent, in the GOCI retrievals
(spectrum 1). However, for a statistically significant result
more direct comparisons between field and satellite data
are required. Therefore, we conclude this section with the need
of further in situ observations for validation purposes.
4. Discussion and Conclusion
The region of the Korean Peninsula, East China and
Yellow Seas is complex for the retrieval and validation of
ocean colour products. A strong limitation is the high
cloudiness of this region which usually impedes ideal
conditions for the atmospheric corrections. Another local
difficulty, both for atmospheric correction and cloud masking, is
the high level of turbidity especially over the Yangtze delta
and close to the coasts of the South-West of Korea. Highly
turbid regions represent a challenge for ocean colour remote
sensing and are ideal cases to assess the current validity of
GOCI radiometric products.
The benefits of the geostationary geometry, as compared
to these sensors, have first been observed with less adjacency
effects and less glint. Although we have not fully exploited
the possibility of acquiring observations every hour over
the same area, the high periodicity of the geostationary
acquisitions allowed a closer temporal coincidence with the
other sensors.
Our analyses show first a relative agreement between
GOCI, MERIS and MODIS seawater reflectance products,
which is quite promising for the exploitation of ocean
colour satellite data in geostationary orbit. Considering
uncertainties in the marine signal for both Sun-synchronous
sensors, and their relative differences, no obvious bias was
found here in the GOCI product. However our results
highlight the need of lowering the cloud detection threshold
and improving the atmospheric correction of GOCI data
over turbid waters. This conclusion is drawn based on
numerous analyses of the reflectance signal prior and after
atmospheric corrections. The problem does not come from
a sensor mis-calibration in the visible wavebands, as
illustrated on Fig. 14 for September 4th and October 4th
2011: on those cases, MERIS and GOCI sensors reveal
strong consistency in Rayleigh corrected signal, in
particular over the Yangtze delta where GOCI nRrs are
unavailable.
Quantitatively, the seawater reflectances retrieved from
MERIS data are typically lower than those retrieved from
GOCI and MODIS. A good agreement is generally
obtained between GOCI and MODIS nRrs products at
wavelengths longer than 443 nm (where calibration issues
are known for MODIS). These results are however
independent from a knowledge of the “ground-truth” which
is the key information to validate the algorithms. The
availability of few in situ data from the KOSC campaign
allowed a comparison of all sensors together against two in
situ high-quality spectra of the nRrs. Although, locally,
22 Lamquin, N. et al.
these two spectra showed better correspondence with
MERIS we cannot but only conclude in the necessity to
gather more of these spectra to obtain statistical confidence
in such comparisons.
We recommend two necessities for the assessment of
GOCI data: 1) the provision of the atmospheric corrected
data over highly turbid waters probably by improving cloud
discrimination over turbid pixels; and 2) the need of numerous
field nRrs measurements over the Yellow and East China
Seas to multiply match-ups with satellite measurements (up
to ten satellite images a day over this region when combining
GOCI, MERIS and MODIS data).
In a short term, systematic delivery of GOCI Rayleigh
corrected reflectance could be a solution for data exploitation
over the most turbid areas.
Fig. 14. Maps of GOCI nRrs and ρRC (first and second row) and MERIS nRrs and ρRC (third and fourth row) at 660 nm. September 4th
2011 (left) and October 4th 2011 (right)
Assessment of GOCI Radiometric Products using MERIS, MODIS and Field Measurements 23
AcknowledgementsWe are strongly grateful to the Korea Ocean Satellite
Center (KOSC)/KORDI team for its help and support in
GOCI data handling. This work was cofunded by the EC
FP7 AQUAMAR project, Centre National d’Etudes Spatiales
(CNES) contract n°116405 and GOYA project (TOSCA,
Principal investigator D. Doxaran). GOCI data were provided
by KOSC/KORDI and processed with the GDPS software,
supported by the Research and Applications of Geostationary
Ocean Color Satellite (PM56890) funded by the Korea
Ministry of Land, Transport and Maritime Affairs (Ryu and
Park). Level 1 MERIS data of the 3rd reprocessing were
provided and processed to Level 2 with the ODESA facility
developed by ESA and ACRI-ST (http://earth.eo.esa.int/
odesa). We thank Julien Demaria from ACRI-ST for his
support on the data reprojection. MODIS data were provided by
NASA and processed at Level 2 with the SeaDAS software
(http://seadas.gsfc.nasa.gov/). We thank two anonymous
reviewers for their fruitful comments.
Fig. 14. Continued
24 Lamquin, N. et al.
References
Ahn, JH, Park YJ, Ryu JH, Lee B, Oh IS (2012) Development of
atmospheric correction algorithm for Geostationary Ocean
Color Image (GOCI). Ocean Sci J (this issue)
Antoine D, Morel A (1999) A multiple scattering algorithm for
atmospheric correction of remotely-sensed ocean colour (MERIS
instrument): principle and implementation for atmospheres
carrying various aerosols including absorbing ones. Int J