Geology Reference
In-Depth Information
the algorithm implements a method to correct for the
influence of the atmosphere. The method involves the
calculation of the expected (modeled) brightness tem-
perature from all SSM/I spectral channels using a for-
ward microwave radiative transfer model developed by
Kummerow
[1993] for different surfaces and atmospheric
conditions. Two pairs of surface types were used to
output model results: FY and thin ice and then FY and
C‐type ice. Twelve atmospheric conditions were selected
ranging from clear skies to winter cumulus cloudy condi-
tions. Each condition was characterized in terms of cloud
type, cloud liquid water contents, base height, and top
height. These conditions characterize different atmos-
pheres in summer and winter seasons in the Arctic.
Figures 10.11c and 10.11d show the model's data (gray
circles) in the ΔGR ‐ PR
R
(19) and ΔGR ‐ PR
R
(85) param-
eter spaces. It can be seen that model results, in general,
span the width of the observed clusters. A clear atmos-
phere has the lowest ΔGR for each surface. More impor-
tantly, variation in atmospheric conditions result in a
wider range of PR
R
(85) than PR
R
(19). The range of
PR
R
(19) is very narrow for open water and almost zero
for the two ice types presented in Figure 10.11c. This
means that PR
R
(19) is a better parameter to resolve total
ice concentration, while PR
R
(85) is a better parameter to
resolve atmospheric conditions. It can also be seen in
Figure 10.11d that PR
R
(85) is also nearly independent
of surface effects, resulting in an almost vertical cluster
of points from all ice types, with or without surface
effects. Nevertheless, the C‐type ice appears radiometri-
cally distinct from other ice types on the ΔGR axis. This
highlights another reason to use PR
R
(85) in the calcula-
tion scheme.
In summary, the rationale behind using the three spec-
tral ratios in the NT2 algorithm is as follows. First, the
rotated polarization ratio from the 19 GHz channel is
used to facilitate discrimination between sea ice and
water. Second, the gradient difference is used to resolve
the ambiguity between pixels with low ice concentration
and pixels with significant surface effects, where observed
brightness temperature is mainly caused by metamor-
phosed snow rather than the ice surface. Third, the
rotated polarization ratio from the 85 GHz channel is
used to avoid ambiguity between changes in ice
concentration and changes in atmospheric conditions
(due to a higher sensitivity of the 85 GHz channel to
atmospheric variability). The calculation scheme of the
NT2 algorithm proceeds as follows (Figure 10.12):
GR GR
GR
VV
(10.21)
85
HH
19
85 19
where PR
R
(19) and PR
R
(85) are called rotated polari-
zation ratios. The parameter ΔGR is called the gradient
difference. Definitions of the rotated polarization
ratios and the angles
ϕ
19
and
ϕ
85
are presented in the
scatterplots shown in Figure 10.11. The rationale for
using the three parameters represented by equations
(10.19)-(10.21) becomes apparent from Figure 10.11.
The figure shows scatterplots of SSM/I data over
Weddell Sea in the Antarctic, acquired on 15 September
1992. In Figure 10.11a the gray circles represent the tie
points of OW as well as ice types A and B, which cor-
respond to FY and MY ice types in the Arctic, respec-
tively. The line A-B represents 100% ice concentration
and the distance from the OW point to that line is a
measure of ice concentration. Label C indicates pixels
with significant surface effects. The cluster of points
between A and C is attributed to surface conditions
such as glaze and layering. They decrease the horizon-
tally polarized brightness temperature and therefore
increase PR
19
. The NT algorithm underestimates ice
concentration in the presence of these conditions since
no provision is included to account for their effects. The
NT2 algorithm uses the rotated polarization ratio,
PR
R
(19) (rather than PR
19
), which results from rotating
the axes in Figure 10.11a by an angle
ϕ
19
, defined as the
angle between the GR axis and the AB line. As such,
PR
R
(19) becomes independent of ice type.
In order to resolve the ambiguity between points of
true ice concentration and points of surface effects, the
gradient difference parameter [equation (10.21)] is used.
Figure 10.11b shows the data distribution in the gradient
space GR
85
H
19
H
and GR
85
V
19
V
. The plot shows a narrow
cluster along line AB.
Markus and Cavalieri
[2000] stated
that points along that line represent 100% ice concentra-
tion with different gradient values corresponding to dif-
ferent ice types. As ice concentration decreases, the two
gradient ratios will have slightly different values and hence
the points start to deviate slightly from line AB. Points
along the curvilinear path BA‐OW also represent lower
ice concentration. When surface effects start to be pro-
nounced, points deviate even further from line AB toward
increased GR
85
H
19
H
. The cloud of points labeled C in
Figure 10.11b corresponds to the cluster of points labeled
C in Figure 10.11a. It is obvious then that the gradient
difference represented by equation (10.21) takes higher
values for ice type C. The algorithm uses a threshold on
ΔGR to designate points of surface effects (i.e., ice type C).
Figure 10.11c shows the location of ice types C with
respect to ice types A and B in the ΔGR, PR
R
(19) space.
Since 85 GHz data are more sensitive to atmospheric
conditions as compared to the lower frequency channels,
1. The forward microwave radiative transfer model is
used to generate a set of modeled SSM/I brightness tem-
perature data for four “pure” surfaces: FY ice, thin ice,
C‐type ice (i.e., ice with surface effects), and open water.
Appropriate emissivity for each surface has to be
