POLinSAR Workshop 2003
Session: Sea Ice
Chair(s): Mark Drinkwater and Wolfgang Dierking
Results were presented from three sensors, from five geographic ice
regimes (both hemispheres)
In spite of this data we are limited when drawing conclusions
about the robustness of classification results.
Only one common dataset was used throughout all presentations
Three different classification approaches have been made to date:
Max. A Posteriori
Hierarchical (knowledge driven)
Unsupervised Wishart together with Alpha/Entropy/Anisotropy
Assessment of class accuracy limited by lack of independent reference
(I.e. in-situ data)
Globally, L-band full polarimetry appears more effective than C-band
for ice classification
Caveats are that regional advantages appear to exist in employing
C-band polarimetric data
The role of phase in polarimetric classification is not clear:
It does however appear to be of some practical value in Open
Water/Ice classification, and/or discrimination of thin ice types
Spatial resolution (spaceborne vs airborne) has as yet an undetermined
effect on polarimetric parameter distributions.
Roundtable Session Questions and Answers
Question 1: What are the applications of sea ice polarimetry (science
Ice centres (NIC; CIS; DMI, etc.) use radar data operationally
in ship routing/logistical & tactical support but image classification
performed largely on wide-swath amplitude images with supervised methods.
· Ocean-atmosphere heat fluxes governed by ice thickesses
Question 2: What applications of sea ice polarimetry are viable based
on anticipated sampling capabilities associated with polarimetric modes?
· Ice thickness governs ice dynamics (i.e. knowledge of distribution
of leads and ridges)
It is as yet undetermined whether swath-width limitations(i.e.
coverage) of future spaceborne polarimetric data will limit their practical
value to operational ice services
· Currently Wide-Swath is the standard product, APmode to be
· Currently used to viewing single-channel intensity images.
It is undetermined whether polarimetry is of any practical benefit
(i.e. added value) to existing operational services.
Currently under evaluation by CIS.
*We must consider the implications of operational (<3hour)
delivery of polarimetric data for established ice services*
Question 3: For classification of sea ice, what accuracies
are we hoping to achieve and what ice types are important to classify -
are we in a position to make a generic statement about these goals?
Operational needs determine that we must:
· discriminate ice/no ice robustly under any wind conditions.
Scientifically: polarimetric data are a means to further understand
the details of e-m interaction with inhomogeneous ice media
· detect thick ice and robustly discriminate between thick and
Near future spaceborne polarimetric operating modes can likely
NOT supplant more standard mode wide swath imaging, due to the time/space
coverage limitations, but rather be used for aperiodic tactical/logistical
support as/when more precise information is available.
Question 4: What are the primary limitations of the existing
work on polarimetric classification of sea ice?
Limitations of existing data sets include:
· Season; region; radar configuration specificity.
A more systematic comparison & evaluation of existing approaches is
needed An independent reference is needed for quantitative evaluation of
Question 5: Which polarimetric parameters or frequencies which
govern our ability to classify sea ice?
No conclusive answer regarding which polarimetric parameters are
indispensible, but to first order (including all data analysed to date)
quad-pol amplitudes appear to govern overall discriminatory and thus classification
Question 6: In what way do the future choices of satellite
frequencies and channels limit our current capability?
No conclusive answer for single frequency polarimetric data until ALOS
and RADARSAT-2 fly
The jury is still out on whether L- or C-band is better overall.
Existing data limitations prevent any judgement,and examples indicate that
C- and L-bands are better under certain circumstances.
For seaice there appear several advantages of multi- frequency
data multi-parameter data over single-frequency polarimetry
Multi-frequency polarimetry is undoubdedly superior to single
Perhaps some opportunities exist for time/space collocation of
PALSAR and RADARSAT-2 polarimetry over sea ice to demonstrate the benefits
Some merit to the idea of tandem operations of TerraSAR-L and
X for multi-freq. data combinations for sea-ice applications.
Question 7: What incremental improvement shall we expect wrt
Envisat alternating polarisation mode? Where are the primary benefits at
ASAR APmode will undoubtedly make an improvement over single channel
Early examples seem to indicate effective discriminatory capability
between thick and thin ice. But the specific advantages over single channel
SAR are yet to be established.
Question 8: What are the recommended actions for future work?
Further studies needed on common datasets which are available
to all study participants.
*This currently limits existing intercomparison of platform-specific
Independent classification references must be established to
quantitatively evaluate performance of different methodologies
Archival material from AIRSAR and EMISAR sea ice data must be
safeguarded for preparations for ALOS and RADARSAT-2
Rather than necessarily acquire new campaign data, new existing
examples should be processed from archived unprocessed airborne polarimetric
Initiate new studies to exploit Earthnet ALOS resource
Sea Ice (Parallel Session) Presentations
Polarimetry of Sea Ice
Mr. Mark Drinkwater
Wishart Classifications of Sea-Ice using Entropy, Alpha and Anisotropy
Polarimetry for Sea Ice Monitoring
strategies for fully polarimetric SAR data of sea ice