Geology Reference
In-Depth Information
bidirectional reflectance from channel 2 and channel 1,
the same ratio between channels 3 and 1, and the tem-
perature difference between the AVHRR channel 4
brightness temperature and a modeled surface tempe-
rature. The Bayesian equation [equation (10.13)] is
extended to combine the four satellite spectral features.
Once again, each pixel is classified to the class with the
highest a posteriori probability. The class density condi-
tional probabilities can be approximated using an appro-
priate standard distribution (e.g., normal or gamma
distribution functions).
Meier
[2005] estimated ice concentration from AVHRR
VIS/IR channels and compared it against estimates from
PM data using four algorithms: BT, Cal/Val, NT, and
NT2 (for designations see section 10.2.2). In the absence
of solar radiation during the winter, ice and water can be
discriminated using surface temperature from a TIR chan-
nel. A threshold of 271 K was selected by the author to
assign a pixel to ice when it is below this temperature. This
method assumes homogeneous pixel contents, which is
not always a true assumption especially at locations near
and within the ice edge. Violation of this assumption
causes an overestimation of ice concentration, but
Emery
et al.
[1991] reported that this error is, in general, small
compared to errors in the SSM/I concentration. During
the summer, on the other hand, the surface temperature of
ice and water become similar. The reflectance from a VIS
channel can then be used not only to discriminate between
the two surfaces but also as an indication of the propor-
tion of water and ice in each pixel [
Meier
, 2005]. This
requires estimation of mean values (tie points) of the
reflectance from each surface using samples of homogene-
ous ice and water pixels. The error in ice concentration
using this approach data was determined to be between
5% and 20% (winter and summer) [
Comiso and Steffen
,
2001] and 6.8%-15.1% (summer) and 8.6%-26.8% (win-
ter) [
Emery et al.
, 1991]. Figure 10.10 presents a case study
from
Meier
[2005] showing ice concentration estimates
from AVHRR (AV) compared to estimates from the afore-
mentioned four passive microwave algorithms. As can be
seen in the original AVHRR channel 2 data (top left
panel), the ice cover is broken with considerable open water
between large and small floes. The white contour delineates
the region of clear sky in the AVHRR image, for which the
concentration can be determined. The AVHRR concentra-
tion field shows less concentration along the right side of
the contour where water is clearly visible in the image.
Other than that the concentration is nearly 100% as can be
visually identified in the original image. The NT algorithm
gives the lowest ice concentration estimate in this summer
scene and the CAL/VAK algorithm gives the highest esti-
mate. Results from Bootstrap and NT2 algorithms are
similar. The advantage of using the finer resolution of
AVHRR is apparent in the figure.
10.2.2. Ice Concentration Using Passive Microwave
Observations
Passive microwave observations are most suitable for
ice concentration estimates for two reasons. First, water
has high dielectric values that translate into low emissiv-
ity and therefore becomes radiometrically colder than
ice. Second, water depolarizes its microwave emission
(Figure 8.17) while ice does not (depolarization in the
PM context means bias toward one polarization; i.e. the
difference between the horizontal and vertical polariza-
tion emissions is large). The second feature is pronounced
at all frequencies as shown in Figure 8.17 and therefore
is used in most ice concentration retrieval algorithms.
However, radiation at higher frequencies is affected by
three atmospheric influences: the integrated water vapor,
cloud water contents, and surface wind over ocean. These
influences should be removed from the observations
before the application of the parameter retrieval algo-
rithm. The advantage of using the high‐frequency chan-
nels, however, is their finer spatial resolution (Table 7.4).
A number of algorithms have been developed to retrieve
ice concentration from PM data. They include the follow-
ing (the list is presented in chronological order but it is
not inclusive): NORSEX [
Svendsen et al
., 1983], NASA
Team (NT) [
Cavalieri et al
., 1984], the University of
Massachusetts — Atmospheric Environment Service
(UMass‐AES) [
Swift et al
., 1985], the Bootstrap algo-
rithm (BS) [
Comiso
, 1986], the Svendsen's near 90 GHz
algorithm (N90GHz) [
Svendsen et al
., 1987], the calibra-
tion‐validation (Cal/Val) [
Ramseier
, 1991], the Bristol
algorithm (BRI) [
Smith and Barrett
, 1994], the Atmospheric
Environment Service — York University (AES‐York)
[
Rubinststein et al
., 1994], the NASA thin ice algorithm
[
Cavalieri
, 1994], the Technical University of Denmark
hybrid (TUD) [
Pedersen
, 1998], the Enhanced NASA
Team (NT2) [
Markus and Cavalieri
, 2000], the Arctic
Radiation and Turbulence Interaction Study (ARTIS)
Sea Ice (ASI) [
Kaleschke et al.
, 2001], SEALION [
Kern
,
2004], and the Environment Canada's Ice Concentration
Extractor (ECICE) [
Shokr et al
., 2008]. Most of the algo-
rithms take advantage of the clustering of the data points
from different ice types and OW in a certain parameter
space. In addition to brightness temperature the parame-
ters that are commonly used are polarization or gradient
ratios as defined in equations (8.10) and (8.11).
While different algorithms vary in details, they can be
grouped into four categories: (1) Solving a set of linear
algebraic equations where each equation decomposes the
radiometric observation into components pertaining to
the OW and ice types within the observation footprint,
weighed by their concentration [see equation (10.15)].
(2) Searching a database of simulated observations gener-
ated for a large number of ice concentrations to find
