Geology Reference
In-Depth Information
that visual analysis of satellite imagery data (especially
from SAR) is still the prime methodology used in opera-
tional ice centers. Sets of rules for image interpretation
have been developed over the years to support this task.
These rules are not covered in this topic.
This chapter addresses methods of the retrieval of
seven parameters: ice types, concentration, area and
extent, thickness, surface temperature, snow depth, and
ice displacement/motion. For some parameters such as
ice concentration mathematical derivation of a few meth-
ods are presented in details. When retrieval is available
from different categories of remote sensors, methods are
grouped based on each category; namely optical, thermal
infrared, passive microwave, and radar. Potential and
limitations of each method or sensor type are discussed.
The chapter includes also sample results from the meth-
ods, comparison between methods, and discussions of
their potential and limitations.
it is not unique to a particular thickness-based ice type.
Similarly, wet ice and snow surface or snow metamor-
phism affect the observed backscatter; yet none of these
features is unique to a particular type. That is why, for
example, when the ice surface melts in the summer the ice
type classification becomes unattainable.
Remote sensing data can be more successful in identify-
ing surface‐based rather than thickness‐based ice types.
For example, they can be successfully used to identify lev-
eled versus deformed ice, ridged versus rafted ice, bare
surface versus snow‐covered ice, smooth versus frost‐
flower‐covered young ice surface, dry versus melt‐ponded
surface, and so on. Nevertheless, efforts have been always
been directed to classify ice according to the standard
thickness‐based WMO‐defined ice types as demanded by
marine operational applications. This is the dilemma of
using remote sensing data to classify ice types. The crite-
rion of ice classification is different than the criteria that
influence the observations.
SAR data is particularly important for ice type classifi-
cation. In fact, the development of the early space‐borne
SAR systems (particularly Radarsat‐1) was mainly driven
by requirements for operational sea ice monitoring,
including classification. At the center of these require-
ments was the identification of navigationally hazardous
types such as MY ice or heavily ridged FY ice. In the
1990s funds were allocated to develop automated or sem-
iautomated SAR‐based systems of ice classification. At
that time SAR images were the prime data source for the
operational ice monitoring programs, but the analysis
was performed visually by expert operators. Summaries
of operational use of SAR data in the Canadian ice mon-
itoring program from the early space‐borne systems
onboard ERS and Radarsat‐1 satellites are presented
in
Raney and Falkingham
[1994],
Shokr et al
. [1996],
and
Ramsay et al
. [1998].
The first attempts to classify ice types in SAR images
adopted common approaches of image analysis tech-
niques. These approaches are presented in several text-
books on remote sensing image analysis [e.g.,
Richards
and Jia
, 2005;
Schowengerdt
, 2006;
Campbell and Wynne
,
2011]. They include supervised and unsupervised clas-
sification techniques, data fusion methods to combine
observations from different sensors, dimensionality
reduction techniques, as well as a few statistical tools.
The accuracy of any approach depends on the probabil-
ity distribution functions of the given radiometric
parameters from the desired ice types. The more overlap
between the distributions the less accurate the classifica-
tion results would be. In this case image classification
may benefit from using contextual information (texture,
shape of ice floes, presence of ridges, etc.) or ancillary
data (ice climatology, wind and temperature fields,
recent history of ice field, etc.).
10.1. Ice Type classIfIcaTIon
The material in this section addresses the classification
of ice types from optical, TIR, and microwave imagery
data. These radiometric observations can be incorpo-
rated in a simple or sophisticated classification technique.
It should be noted, however, that attempts to classify ice
types in support for the operational ice monitoring so far
have not reached a state of maturity that warrants their
incorporation in operational systems, at least in some
centers such as the Canadian Ice Service. Most of the
algorithms that have been developed have failed to fulfill
key operational requirements such as classification accu-
racy, robustness, or fast turnaround. This is particularly
true when using a single‐channel SAR, which is still the
primary data source for operational ice analysis in many
centers. Marine operators prefer to use visual analysis of
near‐real‐time SAR images to identify hazardous ice
types along their navigation routes.
As explained in section 2.6.1, the most popular ice clas-
sification criterion is based on ice age (or equivalently
thickness). This conforms to the criterion used by the
World Meteorological Organization (WMO) to define the
stages of ice development. Since remote sensing observa-
tions are engendered by EM wave interaction with the top
ice or snow layer, the observations can be used only as a
proxy indicator of thickness‐based ice types. This can lead
to successful results if the ice type leaves a distinct charac-
teristic in the top layer. Examples include high salinity of
newly formed ice, smooth surface of Nilas, rafting of
young ice types, frost flowers on new and gray ice surface,
and high porosity of MY ice. Inference of ice type from
SAR data is exceedingly difficult if based on a criterion
shared by a few ice types such as surface roughness. This
is a main factor that influences the radar backscatter, but
