Geology Reference
In-Depth Information
atmospheric interference.
De Abreu et al.
[1994] presented
a correction scheme for the retrieval of albedo over
Arctic ice using data collected during the SIMMS exper-
iment in Resolute Passage, Canadian central Arctic, in the
spring of 1992. The scheme accounts for the intervening
atmosphere, viewing geometry and sensor spectral
response in the VIS range. They found that the atmos-
pheric correction increases the TOA albedo by 27%-32%
(meaning that atmospheric constituents absorb a signifi-
cant amount of radiation in the VIS range). After remov-
ing the effects of viewing geometry, the variability of
surface albedo between orbits decreases. When corrected
for viewing geometry, the satellite‐derived surface albedo
over snow‐covered sea ice ranged from 0.68 to 0.82. Other
atmospheric correction schemes that target a wide range
of applications are presented in
Fraser et al.
[1989],
Rahman and Dedieu
[1994], and
Karpouzli and Malthus
[2003]. The comparison between different schemes is pre-
sented in
Norjamäkiand Tokola
[2007]. Atmospheric cor-
rection that accounts for the scattering, attenuation,
emission, and adjacent effects is addressed in several
topics [e.g.,
Schowengerdt
, 2006;
Kondratyev et al.
, 2012].
Commercial software packages for atmospheric correc-
tions are also available. Two commonly used packages are
the Quick Atmospheric Correction (QUAC) [
Bernstein
et al.
, 2006] and the Fast Line‐of‐Sight Atmospheric
Analysis of Spectral Hypercubes (FLAASH) [
Anderson
et. al.
, 2002], both developed by Spectral Sciences in
United States. The software packages retrieve spectral
reflectance from multispectral and hyperspectral radiance
images. QUAC conducts a more approximate atmospheric
correction (accuracy within ±15%), but at a faster speed
than FLAASH or other physics‐based first‐principle
methods. FLAASH, on the other hand, is a first‐principle
atmospheric correction tool that corrects radiation in
wavelengths in the visible through near‐infrared and short-
wave infrared regions, up to 3
μ
m. Details on the principles
and operation of these two packages of software are
included in
ITT Visual Information Solutions
[2009].
Cloud filtering or masking is another issue that should
be addressed at the preprocessing level of the data. Due
to the higher reflectance of clouds in the visible range
[
Chen et al.,
2002], cloud cover obstructs the sea ice scenes
during most of the daylight season in the polar regions.
On average, 80% of the Arctic Ocean is covered with
clouds. To recover sea ice information from certain area
in these regions the area should be observed for a long
enough period for the clouds to move. Daily observations
would probably be sufficient.
Johannessen et al
. [2007]
recommended receiving images from four to five orbits
per day in order to obtain surface information at least
once in any 3-5 day period. It is worth mentioning in
this context that undetected cloud pixels impede also
the retrieval of atmospheric information (e.g., aerosol
parameters) and render correction for atmospheric influ-
ences difficult.
Numerous techniques have been developed to filter the
cloudy segments from the optical and TIR imagery data,
minimize the cloud effects on surface observations, or just
identify the cloud situation at the pixel level as being fully
cloudy, partially cloudy, or clear. Traditional techniques of
cloud masking for low‐ and medium‐resolution data (e.g.,
AVHRR and MODIS) are usually based on empirically
tuned thresholds from visible and IR channels [
Ackerman
et al.
, 1998;
Dybbroe et al.,
2005]. More sophisticated
approaches that were developed later included the use of
neural networks [
Jang et al.
, 2006], linear unmixing tech-
niques and time series analysis [
Gómez‐Chova et al.,
2007],
and parallel Markovian segmentation [Le Hégarat‐Mascle
and André; 2009]. Cloud masks are part of geophysical
parameter products from operational satellites. The
MODIS cloud mask aims to minimize the effect of cloud
contamination [
Ackerman et al.
, 1998]. It also classifies
each pixel as either confidently cloud, probably cloud, con-
fidently clear, or uncertain.
Yu and Lindsay
[2003] stated
that while various cloud masking techniques are available,
no reliable algorithm has been developed so far for the
Arctic dark season. Visual inspection of the TIR images is
still a usable method for this case.
Schweiger et al.
[2008]
explored connections between cloud cover and sea ice vari-
ability in the marginal ice zone in the Arctic during autumn
using 40 year weather reanalysis from the European Center
for Medium‐Range Weather Forecast (ECMWF) and the
Television and Infrared Observation Satellite (TIROS)
Operational Vertical Sounder (TOVS) Polar Pathfinder
satellite data sets. They found that cloud cover variability is
strongly inked to sea ice variability. Changes in cloud cover
can be explained in terms of the increase of near‐surface
temperatures resulting from the removal of sea ice.
Atmospheric Correction of Passive Microwave
Observations
Due to their long wavelengths relative to
the dimensions of the atmospheric scattering elements,
observations from microwave sensors are not generally
affected by gases or aerosols in the atmosphere at lower
frequencies. Usually, observations from channels lower
than 19 GHz (higher than 1.55 cm wavelength), which
includes frequencies of operational radar remote sens-
ing, can “see” through the atmosphere and clouds pro-
vided that clouds are not precipitating. For passive
microwave frequencies of 37 GHz channel and higher,
observations can be affected by atmospheric constituents
to some degree (depending on the type and size of sus-
pended elements), and therefore a correction must be
applied to recover the microwave emitted radiation from
the surface. The correction should be applied particularly
to the SSM/I 85.5 GHz, AMSR‐E 89.0 GHz, or equiva-
lent channels from future sensors. The atmospheric
