Geology Reference
In-Depth Information
Table 10.2
National centers and institutions that produce ice concentration maps routinely.
Center/Institution
Link
Canadian Ice Service (CIS)
EUMETSAT's Ocean ad Sea Ice Satellite
Application Facility (OSI‐SAF)
Japan Aerospace Exploration Agency (JAXA)
Nansen Environmental and Remote Sensing Center
NASA Goddard Space Flight Center (GSFC)
National Snow and Ice Data Center (U.S.A) (NSIDC)
Norwegian Polar Institute (CliC Int. Project Office)
Technical University of Denmark (PolarView)
University of Bremen (Germany)
Note
: The links to access the data are shown.
In addition to the prime sensor for ice concentration
retrieval (PM), other sensors include VIS, TIR, and SAR.
More focus is placed in this section on retrieval of ice
concentrations from PM data. As mentioned, most algo-
rithms produce total ice concentration, although some
generate partial concentrations of certain ice types.
Concentration of ice types can be useful for certain appli-
cations. For example, concentration of thin ice is impor-
tant for numerical weather prediction and climate models
while concentration of thick ice is more important for
marine navigation purposes.
the unavailability of the VIS data during the dark polar
season and the contamination of both VIS and TIR
observations by clouds and other atmospheric constitu-
ents limit their use. Ice concentration products from these
sensors are generated on an opportunity basis at a few ice
monitoring centers.
AVHRR data from three channels (2, 3, and 4) were
used to produce maps of ice concentration (along with a
suite of other products) from the OSI‐SAF system at the
Norwegian Meteorological Institute (DNMI). Instead of
using a threshold approach to assign ranges of radiomet-
ric measurements to ice and water surfaces, the algorithm
uses a Bayesian approach as described by equation (10.1)
[
Godøy and Eastwood,
2002]. After applying a cloud
mask, the algorithm uses two parameters from AVHRR:
albedo from channel 2 (
A
2
) and the difference of bright-
ness temperatures between channels 3 and 4 (
T
3
−
T
4
). Sea
ice has higher
A
2
and lower
T
4
than OW. When
A
2
is not
available during the dark season, it is replaced by the
difference of TIR observations from channels 4 and
5 (
T
4
−
T
5
). For ice‐OW discrimination using
A
2
and
(
T
3
−
T
4
), equation (10.1) can be rewritten as
10.2.1. Ice Concentration Using VIS and TIR Images
Visible and TIR data (processed separately or com-
bined) are used to estimate ice concentration. As men-
tioned, ice and OW can be identified based on their
difference in albedo/or surface temperature. The advan-
tage of using these data is their relatively finer spatial
resolution compared to the PM data. This reduces
the chances of land contamination and consequently
allows for the estimation of ice concentration in narrow
passages and archipelago regions. On the other hand
pA pT T
(|
ice)
((
)
|
ice) (ice)
p
(10.13)
p ATT
(ice
|
,(
))
2
3
4
2
3
4
pA p
(|
ice)
(
(
TT p
)
|
ice) (ice)
pAp TT p
(|
ow)
((
)
|
ow) (ow)
2
3
4
2
3
4
The a priori probabilities of ice and water
p
(ice) and
p
(ow) in the above equation can be obtained using clima-
tological ice records or can be set to 0.5 in the absence
of this information. The conditional probabilities can be
obtained from pixels of known surface area (training
areas).
Godøy and Eastwood
[2002] found that those prob-
abilities can be approximated by a gamma distribution.
This distribution has the advantages of simulating the
Gaussian distribution around its peak while reproducing
the skewness in the data:
()
(
x
)
1
exp( )
x
/
px
0
0
x
()
(10.14)
where
α
,
β
, and
γ
are distribution parameters and
Γ
is the
gamma function. The parameters were determined for
each month and each surface using training data.
Killie et al.
[2011] extended this approach by using four
AVHRR spectral features for pixel classification:
Channel 1 bidirectional reflectance, the ratio of the
