DataPDF Available

Sea ice drift in the central Arctic combining QuikSCAT and SSM/I sea ice drift data - User's manual V3.0

Authors:
SEA ICE DRIFT IN THE CENTRAL ARCTIC COMBINING
QUIKSCAT AND SSM/I SEA ICE DRIFT DATA
USER’S MANUAL
Version 3.0, April 2008
Fanny Girard-Ardhuin, Robert Ezraty, Denis Croizé-Fillon
and Jean-François Piollé
Laboratoire d’Océanographie Spatiale
Département d’Océanographie Physique et Spatiale
IFREMER, Brest, France
Sea ice drift in the central Arctic combining QuikSCAT & SSM/I sea ice drift data – User’s manual
PolarView Ref. : - C2 – MUT – W – 05 – IF V2.0 2
Acknowledgments
The sea ice drift project was initiated by Alain Cavanié when he was IFREMER Senior
Scientist at Département d’Océanographie Physique et Spatiale/Laboratoire d’Océanographie
Spatiale. He explored, the many ‘bits and pieces’ which paved the way to the operational
algorithm presently in use at IFREMER/CERSAT.
The National Snow and Ice Data Center (NSIDC) provides the gridded Special Sensor
Microwave Imager brightness temperature data and the Physical Oceanographic Distributed
Active Archive Center (PODAAC) provides the L2A level QuikSCAT backscatter data.
The in-situ drifting buoys data set used for validation was extracted from the International
Arctic Buoy Program (IABP) archive.
This project was partly supported by the Global Monitoring for Environment and Security
Initiative of the European Space Agency (PolarView project).
Our thanks to each of these contributors.
Sea ice drift in the central Arctic combining QuikSCAT & SSM/I sea ice drift data – User’s manual
PolarView Ref. : - C2 – MUT – W – 05 – IF V2.0 3
Contents
Revision history ………………………………….………………. 4
Background ………………………………………..…..……. 5
Introduction ………………………………………..…..……. 5
1. The QuikSCAT and SSM/I drift data sets ……….……….…..……. 5
1.1. The composite QuikSCAT backscatter map ..……..…....……. 6
1.2 the SSM/I brightness temperature maps …..………………..……. 7
2. Why a merged drift map ? …………………………………….……. 7
3. The merging process ………..……………………….…..……. 8
3.1. The initialization phase ……….…………………………....……. 9
3.2 Drift vector selection among likely solutions ……………..……. 12
3.3 Validation procedures ……………………………………..……. 14
3.3.1 The difference scheme .…..……………….………..……. 14
3.3.2 The median filter scheme ……………………………..……. 14
4. Analysis of the merged drift product ..…………………………………. 15
5. Data access & organization ..………….………………………..……. 15
5.1 Data access …………………………………...………..……. 15
5.2 Data …………………...………………………..……. 17
5.3 Quicklooks …………………...………………………..……. 19
6. Validation of the Merged drift data with in situ buoys ……………..……. 19
References ………..……………………………………………..……. 21
Contact ……………..………………………………………..……. 23
Annex …………………….………………………………..……. 24
Sea ice drift in the central Arctic combining QuikSCAT & SSM/I sea ice drift data – User’s manual
PolarView Ref. : - C2 – MUT – W – 05 – IF V2.0 4
Revision history
Version Name Date Comments
Draft 0 Ezraty 01/02/2004
Draft 0.1 Piollé 03/03/2004 Section 5 added
V1.0 Ezraty 01/04/2004
V2.0
Girard-
Ardhuin
10/02/2006 Upgrade incomplete daily dataset
Monthly drift maps
New quicklooks
New data access
Section 6 added (validation)
V2.0 Girard-
Ardhuin
20/04/2006 New quicklooks path
V2.0 Girard-
Ardhuin
02/02/2007 Upgrade incomplete daily dataset
V3.0 Girard-
Ardhuin
02/04/2008 New missing values filled dataset (3 and 30
days)
Sea ice drift in the central Arctic combining QuikSCAT & SSM/I sea ice drift data – User’s manual
PolarView Ref. : - C2 – MUT – W – 05 – IF V2.0 5
Background
In situ observation of sea ice motion relies on a limited number of drifters which transmit their
position but also meteorological parameters (http://iabp.apl.washington.edu/).
In complement to this Lagrangian approach, sea ice motion has been estimated over both polar oceans
from microwave brightness temperature data obtained from space borne radiometers. Different pre-
processing algorithms have been used although the core of the main processing relies on correlations
between time-lagged data. Typical examples can be found in Kwok et al., 1998; Liu and Cavalieri
1998; Martin and Augstein, 2000. Similar correlation techniques have also been used for data from
active sensors, for example the ERS-1 scatterometer (Gohin et al., 1998), the NASA scatterometer
NSCAT (Liu et al., 1999) and recently the QuikSCAT scatterometer (Ezraty and Piollé, 2001).
Introduction
As soon as sea ice drift estimates had been produced from scatterometer backscatter maps, the need
for validation of these new drift data sets involved comparisons with in-situ drifting buoy data and
SSM/I derived drift data. The idea of merging the various drifts arose rapidly in order to increase the
density of the drift estimates (see for example Liu et al., 1999). The merging procedures proposed
were quite coarse, relying on weighted-block averaged of the available drift vectors. For example in
the previously quoted publication, weight of 50% for the buoys, 25% for NSCAT and 25% for SSM/I
were used. Other weighs selections are possible, and were also used by other authors, for example
weights proportional to the inverse of the noise levels of the buoy data, the backscatter and brightness
temperature maps.
Our approach is different. The merged drift map relies on the selection and/or the validation, at each
grid point, of a single drift vector among the solutions provided (or partly provided) by the set of the
three drift maps, namely the QuikSCAT drift vectors (hereafter label as Q), the SSM/I drift vectors at
horizontal polarisation (hereafter label as H) and the SSM/I drift vectors at vertical polarisation
(hereafter label as V).
This report is composed of 6 sections.
The QuikSCAT and SSM/I (H and V) individual drift data are presented, in short, in Section 1. The
justifications for a merged drift map are presented in Section 2. The overall processing scheme is
presented in Section 3 where each step is detailed. An analysis of the merged drift data set is presented
in Section 4 and Section 5 presents the information on data access and data organisation. Section 6
presents the validation with buoys data.
1. The QuikSCAT and SSM/I drift data sets
The core of the drift processing algorithm of the QuikSCAT backscatter map and of the SSM/I
brightness temperature maps, the internal validations of each drift product, the quality flag values and
the output file structures are identical for the three data sets. Of course, the geographic coverage and
the drift grid spacing (62.5 km x 62.5 km) remain unchanged. The version 2.0 (January 2004) of the
Sea ice drift in the central Arctic combining QuikSCAT & SSM/I sea ice drift data – User’s manual
PolarView Ref. : - C2 – MUT – W – 05 – IF V2.0 6
drift algorithm is used. A detailed description of these common/similar features can be found in Ezraty
et al., 2006b.
Basically , a template matching procedure (estimated through a standard correlation computation) is
applied to time lagged (3 and 6-day lags) maps of the Laplacian field of either the backscatter map or
the brightness temperature maps. A quality flag is attached to each drift vector, positive values
indicate a validated/likely drift estimation; the different negative values flag the various cases when
the drift estimation should be dismissed. These internal controls consist of:
- the verification of the local ice-drift direction with respect to the interpolated ECMWF surface wind
direction
- local sea-ice drift field consistency tests based on statistical values estimated from the surrounding
ice-drift vectors.
The major difference between the QuikSCAT and the SSM/I drift processing are at the pre-processing
level in order to shape the backscatter and the Tb maps. These maps are built on the 12.5 km x 12.5
km polar stereo projection used by the
National Snow and Ice Data Center (NSIDC), Boulder,
Colorado to disseminate the 85 GHz brightness temperature maps.
1.1 The composite QuikSCAT backscatter map
Since 1999, daily QuikSCAT backscatter maps at 46° (H polarization channel) and 56°(V polarization
channel) incidence angle are produced at CERSAT over the polar regions (Ezraty and Piollé, 2001).
From an earlier study, using the NSCAT scatterometer which operates at the same frequency band as
QuikSCAT, it was shown (Ezraty and Cavanié, 1999) that, over sea ice and at the same incidence
angle, the ratio of V polarization backscatter to H-pol. backscatter is almost constant and equal to one,
and is practically independent of ice type. Thus, the backscatter difference in backscatter between the
two QuikSCAT channels can be mainly attributed to the incidence angle difference. Since this angular
difference remains constant within less than about 1.5°, it makes it possible to estimate most of the
data gap, at the North Pole, in the H-pol. backscatter map from the V polarization backscatter map (Cf.
Plate 1 in Ezraty and Piollé, 2001) which, given the sensor geometry, has a much smaller geographic
data gap. A composite backscatter map is then constructed as half the sum of the modified H
polarization and V polarization maps. The remaining small data gap (< 26 pixels) is filled in using a
quadratic surface interpolation using the surrounding pixels.
The open-water/sea ice discrimination is performed using the QRAD processing (Ezraty et al. 2000).
The benefits from such a pre-processing is three-fold:
- first the whole scatterometer processing, from building the backscatter map up to the drift
computations relies on QuikSCAT data only.
- second, around the North Pole, it avoids the discontinuities which cause systematic and permanent
artefacts when estimating the Laplacian field and thus creating a ‘strange attractor’ effect when
computing the correlations between lagged templates.
- third, averaging two maps decreases the standard deviation of the noise level of the pixels by
2
.
This improvement is really important since a double spatial derivative is to be used at the next step of
the processing. This is the main reason of our confidence in the drifts estimated from QuikSCAT
compared to those derived from the SSM/I .
Sea ice drift in the central Arctic combining QuikSCAT & SSM/I sea ice drift data – User’s manual
PolarView Ref. : - C2 – MUT – W – 05 – IF V2.0 7
1.2 The SSM/I brightness temperature maps
The daily SSM/I 85 GHz brightness temperature maps are obtained either from the CD-rom
distributed by the NSIDC when processing historical data (prior to winter 2003-2004) or from their
web site, for near real time processing, at:
ftp://sidads.colorado.edu/pub/DATASETS/PASSIVE_MICROWAVE/POLAR_STEREO/DATA/TB/
F13/NEAR_REAL_TIME/
The H polarization and V polarization 85 GHz brightness temperature maps are processed separately
since each channel responds to different physical properties via their different emissivity. Moreover,
because of their different sensitivity to physical surface temperature changes and to precipitations it is
not possible to combine them in a single composite map.
The wide data gap at the North Pole is replaced by a fictive land area. The open-water/ice boundary is
estimated from the sea-ice concentration map computed with the ASI sea-ice concentration algorithm
(Kaleschke et al. 2001) which uses both the low (19, 22 and 37 GHz) and high (85 GHz) frequencies
channels of the SSM/I. Since a weather filter is applied, the no-ice area boundary corresponds to about
15% sea ice concentration. Because the 85 GHz channels are sensitive to atmospheric effects, most of
the large spurious sea-ice areas, in the open ocean, are dismissed using a monthly dependent
climatological sea ice mask The few small areas where sea ice artifacts might remain are not a
problem because, being due to short-lived atmospheric effects, theirs impact cancel out when
computing three-day or six-day lagged correlations. At last, in order to dismiss near-shore “land-
contaminated” pixels, the land masks is expanded by two pixels (25 km).
2. Why a merged drift map ?
When validating the scatterometer derived drift vectors with SSM/I derived drift vectors (Carlut,
2003), it was observed that the amount of drift vector pairs in the scatter-plots of the vector
components was lower than about 15% compared to the amount of available drift data from each
sensor. To illustrate this point, Figure 1 presents the time series, during winter 1999-2000, of the
density of drift vectors (as a fraction of the number of drift grid nodes where sea ice is present) for
each sensor. In this plot, the number of drift vectors obtained as “raw” composite is the result of filling
in the gaps of the scatterometer validated drift vector map by the SSM/I H or V estimated drifts when
available and validated. No selection among possible different solutions or further validation is
performed.
This figure clearly confirms the complementarities and the benefits of merging the drift vectors
obtained from QuikSCAT and SSM/I H polarization and V polarization data. The number of “raw
composite” drift data are very often above 90% while QuikSCAT and SSM/I drifts only reach 75 to
80% of the possible nodes of the drift grid. The SSM/I V-pol. channel is slightly more efficient than
the H-pol. channel but both provide very low scores during the freeze-up period (from September 1
st
until mid-October) and from the beginning of the melting period (beginning of May). Because of the
sensitivity of the 85 GHz channels to atmospheric events, large synchronised fluctuations occur (more
than 20 %) in their score.
As soon as the end of September the QuikSCAT score reaches 70% and increases steadily up to
slightly over 80% at the end of the winter period. It is only by mid-May that melting affects its score.
Sea ice drift in the central Arctic combining QuikSCAT & SSM/I sea ice drift data – User’s manual
PolarView Ref. : - C2 – MUT – W – 05 – IF V2.0 8
The North pole data gap in the SSM/I brightness temperature data induces only a ten’s of missing drift
vectors. As regards the comparison of the scatterometer and radiometer efficiency (and certainly not
the geophysical information), this amount can be neglected as can be seen on Figure 1 where the
Scatterometer and SSM/I score are very similar during the core of the winter season.
The score of the raw composite remains quite steady during the cold season since the fluctuations of
the QuikSCAT and the SSM/I scores are in phase opposition.
Building up a merged drift data set makes sense only if each of the primary data sets have similar
statistical properties. This was verified, over the winter periods 1999-2000, 2000-2001 and 2001-2002
up to the fourth moment of the distributions of the drift-vector components. In particular the third and
fourth moment, very sensitive respectively to the asymmetry and to the number of large values were
found to be quite similar.
Data from drifting buoys will not be incorporated in this merged product, as was done by Liu and al.
(1999). A typical daily winter-time drift map over the central Arctic is composed of about 1600
validated drift vectors, for a total of about 2000 grid nodes. The few tens of buoys regularly operated
in the Arctic will not add a significant number of drift-vectors. These drift-vectors do not follow a
statistical law similar to that of the satellite derived data. Furthermore, these data should be
interpolated to the nearest grid node, thus decreasing their impact. Last, they would have rendered
more complex the combination process by introducing an heterogeneous data set (data files access,
data validation, merging algorithm).
Oct
1999 Nov
1999 Dec
1999 Jan
2000 Feb
2000 Mar
2000 Apr
2000 May
2000
0.0
0.2
0.4
0.6
0.8
1.0
Drift data density
3-day lag SSMI-H
SSMI-V
QuikSCAT
Raw Composite
Figure 1 : Time series of drift data density for QuikSCAT, the SSM/I channels and for
the raw composite (winter 1999-2000).
3. The merging process
The core of the merging process relies on previously validated drift data of the merged drift field (the
initialization of the process is described in Section 3.1). The analysis of possible combinations, at a
Sea ice drift in the central Arctic combining QuikSCAT & SSM/I sea ice drift data – User’s manual
PolarView Ref. : - C2 – MUT – W – 05 – IF V2.0 9
single grid-node, indicates 14 possible cases ranging from a triplet of vectors to a single vector. They
are described in Table 1.
Description of combinations
Triplets
Q=H=V Q=H≠V
Q=V≠H
H=V≠Q
Q≠H≠V
Pairs Q=H Q=V H=V Q≠H Q ≠V H≠V
Single Q H V
Table 1 : Possible combinations of drift vectors at a single drift-grid node.
In the following, the cases where a selection among the possible drift-vectors (triplets and pairs) is to
be made are treated differently from the cases where a validation must be performed (single vector).
Both procedures use the surrounding and validated drift vectors of the merged drift field. The selection
procedure is based on the performance of a cost-function while the validation procedure relies on a
comparison with the already validated surrounding vectors. It must be point-out that this selection
procedure followed by the validation process depend on the location of the starting grid-node to be
validated. To circumvent this drawback, a loop is performed until no new vector is accepted as shown
in the block diagram of the merging algorithm presented in Figure 2.
The special validation case when a drift vector is totally isolated from already validated vectors is
treated specifically at the end of the merging process. Here also, a separate loop is performed until no
new vector is validated This case occurs quite often at the edges of the drift-grid; in this case the
vicinity matrix is adapted to the edge.
A flag value is attached to each accepted drift vector, detailing the selection and validation criteria (see
Section 5.2).
The drift data sets, on January 7
th
2003, are used to illustrate the various steps of the algorithm. The
scatterometer drift data is presented in figure 3, the SSM/I drift data is shown in figure 4 and the result
of the merged is illustrated in figure 5.
3.1 The initialization phase
After sorting the data according to their location on the drift grid-nodes, the backbone of the merged-
map is built at each drift grid-nodes where triplets and pairs of identical vectors are found. It is very
unlikely that an outlier value occurs at the same grid point for a pair of identical vectors and a fortiori
for a triplet. Accounting for the “better” quality of the composite backscatter map, the identical triplets
drifts are given a weight of one, identical pairs (including those within a triplet) containing a
QuikSCAT vector are given a weight of two, identical pairs (including those within a triplet) of SSM/I
V polarization and H polarization drift data are given a weight of three. Higher the confidence in the
drift data, the less weight is given. These weights are used in the following selection process. Identical
triplets of drift vectors and pairs are drawn respectively in red and green in figure 5.
Sea ice drift in the central Arctic combining QuikSCAT & SSM/I sea ice drift data – User’s manual
PolarView Ref. : - C2 – MUT – W – 05 – IF V2.0 10
Figure 2 : Block diagram of the merging algorithm
Validation of single drift vector
Areal consistency using only the merged (accepted) surrounding drifts
Yes No
Save Merged drift file
END
Has the number of
valid vector changed ?
SSM/I V
Drift File
SSM/I H
Drift File
QuikSCAT
Drift File
Inventory and sorting of existing combinations (14 cases, Cf. Table 1)
Building the backbone of the Merged drift map from
Identical triplets and pairs of drift vectors
Drift vector selection for multiple solution cases
Cost function scheme
Yes No
Selection or Validation of non rejected remaining drifts; Edge points processing
Median filter scheme (all available channels of neighbouring drift data is used)
Has the number of
valid vector changed ?
Sea ice drift in the central Arctic combining QuikSCAT & SSM/I sea ice drift data – User’s manual
PolarView Ref. : - C2 – MUT – W – 05 – IF V2.0 11
Figure 3: Drift vectors at 3-day lag computed from QuikSCAT backscatter maps (VV polarization),
January 7
th
-10
th
2003). Drift vectors less than one pixel are marked with a cross.
Figure 4 : Drift vectors at 3-day lag computed from SSM/I 85 GHz brightness temperature maps
(January 7
th
-10
th
2003). Identical drift for H and V channel, V channel, H channel. Drift vectors
less than one pixel are marked with a cross.
Sea ice drift in the central Arctic combining QuikSCAT & SSM/I sea ice drift data – User’s manual
PolarView Ref. : - C2 – MUT – W – 05 – IF V2.0 12
Figure 5 : Merged drift vectors at 3-day lag computed from SSM/I & QuikSCAT data (January
7
th
-10
th
2003). Identical drift for QuikSCAT, SSM/I H and V polarizations, identical drift for any
of 2 products, selection or validation of one product. Drift vectors less than one pixel are marked
with a cross.
3.2 Drift vector selection among likely solutions
In the four possible combinations where two or three drift-vector values occur at the same grid-node
(Cf. Table 1), the ambiguity removal relies on a cost function. The selected solution is the vector
solution of :
(
)
321
,,
===
=
kkkol
SSSMINS
or
(
)
21
,
==
=
kkol
SSMINS
depending on the case considered (triplet or
pair)
with:
( ) ( )
[
]
22
24
1
kiki
n
i
i
kxxyyS +=
=
λ
Where:
ii yx
,
are the components of the validated vectors of the merged drift-grid surrounding the
considered grid-node (5 x 5 matrix),
kk yx
,
are the components of the vectors to be tested and
i
λ
are the cost factors attached to each test
geometry.
Sea ice drift in the central Arctic combining QuikSCAT & SSM/I sea ice drift data – User’s manual
PolarView Ref. : - C2 – MUT – W – 05 – IF V2.0 13
The cost factor,
i
λ
, is a function, both, of the weight,
i
ω
, attached to each drift vector of the merged
map and of the distance,
i
d
, between the grid-node of the vector being tested and the grid-nodes of the
surrounding vectors. It is defined as:
iii
d.
ωλ
=
1.0 1.5 2.0 2.5 3.0
Distance (pixels)
0
5
10
15
20
Cost factor
Figure 6 : Cost factor variations as a function of weights and distances.
The weight is a fixed value depending on the selection/validation procedure and on the selected
sensor/channel, while the distance will vary when a given grid-node contributes to another selection
geometry.
For a distance of one pixel (figure 6), the cost factor equals the weight.
The weight values of one, two and three have been introduced in section 3.1 at the initialization step.
Weights values of five and six correspond to either a drift selection among possible vector values or to
a validation of single values using only the surrounding vectors of the merged map. The weight of five
is attributed to a QuikSCAT selected or validated vector while the weight of six is attributed to a
SSM/I vector.
As regards the contribution of the different vectors to the cost function (location and quality of the
vectors for a given geometry), figure 6 shows that a triplet value, even at the farthest position,
contributes almost as much as a “near” pair value, to the minimization of the cost function. The same
applies to a “ far” pair compared to a “near” QuikSCAT selected vector. This choice of cost factor
values emphasizes the impact of the most likely backbone values of the merged map. The maximum
impact (minimum cost function) is given to QuikSCAT vectors.
Other possibilities exist for shaping the cost factor. The choice made here is one of the simplest
solutions and proved to be robust and very efficient.
Sea ice drift in the central Arctic combining QuikSCAT & SSM/I sea ice drift data – User’s manual
PolarView Ref. : - C2 – MUT – W – 05 – IF V2.0 14
3.3 The validation procedures
At a given grid-node, a single drift vector is to be validated/accepted by comparison to the surrounding
drift-vectors. Two cases are considered depending on the drift map used as background: either only the
already accepted vectors of the merged map are used (Difference scheme) or the background is a
composite built from the merged map and from the not-yet accepted vectors from the three drift maps
( Median filter scheme).
3.3.1 The difference scheme
This algorithm assumes that the single vector to be validated is at the center of a 5 x 5 matrix of
vectors of the merged drift-grid containing at least four (accepted) vectors. The statistical study on the
differences between the QuikSCAT and SSM/I valid drift components indicates that the standard
deviation of differences is half a pixel (Carlut,2003) . Thus, the criteria used for validation is :
2+
àmom
yyxx
where m
x
and m
y
are the components (in pixel units) of the mean of the drift-vectors surrounding the
vector to be tested the components of which are
0
x
and à
y
(in pixel units). This criteria enables either
each component to differ by, at most, one pixel or a single component to differ, at most, by two pixels.
The vectors which do not fulfil this condition are dismissed.
A verification is performed on the angular difference between the mean vectors and the accepted
vectors.
3.3.2 The median filter scheme
The goal of this final validation steep is to recover the maximum number of likely drift vectors
acknowledging that they have been “validated” at the individual drift processing stage, but that a very
few number of outlier vectors may remain, mainly at discontinuities or at the edges of the drift-grid.
The standard local geometry remains the 5 x 5 grid-node area although at the edges, this local area
reduced to a 4 x 5 dissymmetric areas. In these areas, the components of the reference vector is built as
the median values of the components of all available vectors either from the merged map or from the
three initial drift maps. To be meaningful, the median filtering technique requires enough vectors in
the sample. From trial and error tests, a minimum of seven vectors was found to be a reasonable
number. In order to improve the efficiency of this technique and in order to increase the impact of the
already validated vectors in the sample, these vectors are counted twice. The same difference criteria
as in Section 3.3.1. is applied between the reference median vector and the vector to be tested. Here
also, the validation relies on the starting location thus, as shown in the block diagram of the algorithm
(figure 2), a loop is performed until no new vector is accepted.
A example of the usefulness of this procedure can be seen when comparing the SSM/I derived drifts
(figure 4) to the QuikSCAT derived drifts (figure 3) in the Canadian Archipelago and North of Baffin
Bay. The algorithm dismisses the non homogeneous set of SSM/I drifts and selects the QuikSCAT
drifts.
Sea ice drift in the central Arctic combining QuikSCAT & SSM/I sea ice drift data – User’s manual
PolarView Ref. : - C2 – MUT – W – 05 – IF V2.0 15
4. Analysis of the merged drift product
In one way or another, the merged drift product relies on the local consistency of the drift-vector field.
The typical coherence scale is a 250 km x 250 km area. When not enough data is available, to validate
a given vector, this vector is “lost” for the merged product but remains available in the individual drift
maps. A typical number for such losses would be twenty vectors. Merging the three data set not only
provides a extra quality control, filtering unlikely outlier values, on each estimated drift vector, it also
enables a denser number of drift vector for the merged map as can be seen on figure 1 and Annex.
These plots confirm the increase, by 12% to 15%, in drift data density anticipated in Section 2. The
slight difference between the 3-day and the 6-day curves are due to the “memory loss” of the sea ice
surface properties as the time lag increases. This is particularly evident when sudden atmospheric
perturbations (wet precipitation) blur the backscatter and radiation signatures, for example in early
December 2002. Also, it must be remembered that the correlation technique used detects only
translations of the patterns and not the rotations which likely occur as the time lag increases and at
locations where free drift is possible (marginal ice zones).
As anticipated, the benefit of combining scatterometer and radiometer derived drifts is also evident
before and after the freezing period. Yet, at the end of the winter period, by April 30
th
May 10
th
, it
was occasionally observed that some of the estimated drift vectors could not be attributed anymore to
sea ice but rather to the signature, on the surface, of the slow spring warming which produces local
and persistent patterns (large melting areas) advected by the wind regime and who are detected by the
correlation computations.
5. Data access & organization
5.1 Data access
Ice motion is estimated during typical winter periods which start on October 1
st
, and ends on April
30
th
, the following year. Depending on the winter period considered, daily drift estimates can be
computed a few days prior or after these dates. These data are not included in the present product in
order to maintain the data set homogeneity but they are available on request.
Monthly drift
Monthly drifts are estimated from the daily drift data (example in figure 7). This data set is basically
built summing, at each grid point, 5 consecutive 6-day lagged drifts of the Merged product starting on
the first day of each month. When a 6-day lagged drift is not available or if a 6-day drift is greater than
62.5 km (5 pixels), two 3-day lagged periods are used. If a 6-day drift is not available for the Merged
product, the QuikSCAT drift product (or eventually the SSM/I drift as an ultimate choice) is used.
3-daily missing values filled drift
This dataset is based on spatial and temporal interpolation of the 3-daily Merged drift during a full
winter.
Sea ice drift in the central Arctic combining QuikSCAT & SSM/I sea ice drift data – User’s manual
PolarView Ref. : - C2 – MUT – W – 05 – IF V2.0 16
30-days missing values filled drift
Files are built like the 30-days or monthly drift dataset, except that when a 6-day lagged drift is not
available or if a 6-day drift is greater than 62.5 km (5 pixels), 3-daily missing values filled drift are
used.
The datasets are available at the following ftp address
ftp://ftp.ifremer.fr/ifremer/cersat/products/gridded/psi-drift/data/arctic/merged-quikscat-ssmi/
The psi-drift folder contains the sea-ice drift motion vectors, documentation and quicklooks. It is
organized as follow :
Sea ice drift in the central Arctic combining QuikSCAT & SSM/I sea ice drift data – User’s manual
PolarView Ref. : - C2 – MUT – W – 05 – IF V2.0 17
Figure 7 : Merged monthly drift vectors computed from SSM/I & QuikSCAT data (March 2005).
Drift vectors less than one pixel are marked with a cross.
5.2 Data
The sea-ice drift vectors are computed over 3 days and 6 days respectively, estimation of monthly
drifts are also available. The data are stored in two different format.
ASCII format
Each file, contains the list of all retrieved vectors. Columns corresponds to the following parameters :
<vector index>, <latitude of start point>, <longitude of start point>, <latitude of end point>,
<longitude of end point>, <quality flag>.
NetCDF format
Each vector (given by its origin latitude-longitude location, zonal and meridional components, and its
quality flag) is mapped over the NSIDC grid. The grid is stored in a NetCDF file. Detailed information
on this format is available at http://www.unidata.ucar.edu/packages/netcdf/
The NetCDF format is self-explanatory and contains the variable names, scaling factors and units. It
must be noted that the quality flag values are coded in signed byte.
Sea ice drift in the central Arctic combining QuikSCAT & SSM/I sea ice drift data – User’s manual
PolarView Ref. : - C2 – MUT – W – 05 – IF V2.0 18
Quality flag
The purpose of the flag descriptor is to indicate which sensor/channel solution was used to built-up the
merged map and how this selection or validation was performed. The flag is built as a two-field byte.
Bit 0 to bit 2 indicate which channel is (are) accepted and bit 3 to bit 6 indicate the selection or
validation criteria employed. Bit 7 is not used.
The flag descriptor is stored on a byte. The significance of each bit (when set to 1) is as follow :
Daily drift
bit description
0 QuikSCAT vector was selected as drift vector solution
1 SSM/I H-pol vector was selected as drift vector solution
2 SSM/I V-pol vector was selected as drift vector solution
3 Two or more drift vectors were identical in the source maps and therefore
selected as the merged drift vector solution (refer to bit0, bit1 and bit2 to check
which ones)
4 The cost function criteria was used to select among all possible solutions
5 Validation of a single vector solution using the accepted neighbouring drift
vectors of the merged map (Local consistency)
6 Validation of a single vector solution using all available information, including
not yet accepted drift vectors (Median filter)
7 spare
Monthly drift
bit description
1 Valid drift vector
Negative values
(-1 to –4)
Number of missing 6-day periods
3-daily and 30 days missing values filled drift
values description
1 Valid drift vector
0 Not valid
Unavailable SSM/I or QuikSCAT data induce data gaps in the time series of daily concentration maps.
If a SSM/I drift is not available for the Merged product, the Merged drift is missing. If a QuikSCAT
drift is not available, the two SSM/I channels are used to compose the Merged product (examples from
1991 until 1999).
The distinction is done between Merged unavailable data (m : missing) and Merged partial-coverage
data (c) (see Annex).
Sea ice drift in the central Arctic combining QuikSCAT & SSM/I sea ice drift data – User’s manual
PolarView Ref. : - C2 – MUT – W – 05 – IF V2.0 19
5.3 Quicklooks
Quicklook pictures of Merged drifts (example in figures 5 and 7) are available at
ftp://ftp.ifremer.fr/ifremer/cersat/products/gridded/psi-drift/quicklooks/arctic/merged-quikscat-ssmi/
6. Validation of the Merged drift data with in situ buoys
The results presented here are extracted from a validation study performed by Girard-Ardhuin (2005).
The validation of the Merged product at 3 and 6-day lags is presented in comparison with drifting
buoys of the International Arctic Buoy Programme (IABP) over five winters (1999-2004).
Figure 8 shows a 3 day lag Merged and buoys drifts with probability in logarithmic scale, showing a
slight dissymmetry towards the strong values for the buoys drifts. The drift values comparison can be
quantified by the standard deviation of the difference (Merged-buoys), which is 7,5
±
0,1 km of which
5,1 km are uncorrelated noise. The quantification effect accounts for
2
δ
/6 with
δ
the pixel size (in
variances), in this case this corresponds to 5,1 km, the uncorrelated noise is thus totally explained by
the quantification effect (Ezraty et al., 2007b).
In speed values, the standard deviation of the difference of ice speeds is 2,91
±
0,04 cm/s, which is
comparable with standard deviation of Liu et al. (1999) with SSM/I and NSCAT at 4 day lag (2,96
cm/s and 2,80 cm/s respectively), or 2,6 to 2,9 cm/s at 1 day lag of Liu and Cavalieri (1998).
Angles of Merged drift present a strong uncertainty : for small drift (lower than two pixels) the angle
is 0, 45 or 90° whereas angles of buoys drift vectors are more accurate. The standard deviation of the
difference is 39,2
±
0,5° of which 33,5° are uncorrelated noise. It decreases down to 29,4
±
0,4° if drift
less than one pixel are excluded.
Figure 9 shows the angle difference between the Merged product and buoys as a function of Merged
drift : the angle difference sharply decreases (smaller than 45°) for drifts higher than 40 km (about 3
pixels), this was also noticed by Liu et al. (1999).
Using 6 day lag is more adequate to small drift, with a better angular resolution with a standard
deviation of the difference of 29,6
±
0,4°.
Figure 8 : Comparison of buoys and Merged drifts for 3 day-lag,
winters 1999-2004. Colors are probability expressed in logarithm
scale (Girard-Ardhuin, 2005).
Sea ice drift in the central Arctic combining QuikSCAT & SSM/I sea ice drift data – User’s manual
PolarView Ref. : - C2 – MUT – W – 05 – IF V2.0 20
Figure 9 : Angle difference between Merged and buoys as a function of
Merged drift for 3 day-lag, winters 1999-2004. Colors are probability
expressed in logarithm scale (Girard-Ardhuin, 2005).
Comparing buoys and Merged drifts in North/East components frames enables independent estimate
of uncertainties. North and East components standard deviation of the difference are 7,0
±
0,1 km of
which 4,0 km and 4,3 km respectively are uncorrelated noise. For each component, the quantification
effect accounts for
2
δ
/12, thus 3,6 km of the uncorrelated noise are explained by this effect.
The Merged drift is validated with buoys drifts over five winters. Merged drifts at 3 day-lag are
constraint by the ability to measure small drifts, the 6 day-lag is more adapted to small drifts but is
constraint by the 6 pixels maximum drift set in the algorithm. The day-lag must be chose in function
of the value of the drifts (Girard-Ardhuin, 2005).
See also the AMSR drift product and the comparison with Merged drift in Ezraty et al., 2007c.
Sea ice drift in the central Arctic combining QuikSCAT & SSM/I sea ice drift data – User’s manual
PolarView Ref. : - C2 – MUT – W – 05 – IF V2.0 21
References
Girard-Ardhuin, F., Dynamique de la banquise de l’Océan Glacial Arctique. Rapport d’avancement
des travaux. (in French), Rapport Scientifique DOPS/LOS - CNES/IFREMER, Avril 2005.
Carlut, C., Validation, critique et fusion de champs de dérive des glaces de mer en Arctique. Rapport
de projet de fin d’étude. Rapport IFREMER/DOPS/LOS 2003/03 (in French), 2003
Ezraty R. and A. Cavanié, Construction and Evaluation of 12.5km grid NSCAT backscatter maps over
Arctic sea ice, IEEE Transactions on Geoscience and Remote Sensing, Vol. 37, no. 3, pp. 1,685-1,697,
May 1999
Ezraty R, W. L. Jones and J. Zec, On the joint use of QSCAT and QRAD over the Arctic and
Antarctic Oceans. Proceeding. IEEE IGARSS 2000, Honolulu, Hawaii, USA, 24-28 July 2000, pp.
1,223 -1,225.
Ezraty R.,F. Girard-Ardhuin, J. F Piollé, L. Kaleschke and G. Heygster, Arctic and Antarctic sea-ice
concentration and Arctic sea ice drift estimated from Special Sensor Microwave Imager data. User’s
manual, Version 2.1, February 2007a.
ftp://ftp.ifremer.fr/ifremer/cersat/products/gridded/psi-drift/documentation/ssmi.pdf
Ezraty R., F. Girard-Ardhuin and J. F. Piollé, Sea-ice drift in the Central Arctic estimated from
SeaWinds/QuikSCAT backscatter maps. User’s Manual, Version 2.2, February 2007b.
ftp://ftp.ifremer.fr/ifremer/cersat/products/gridded/psi-drift/documentation/quikscat.pdf
Ezraty R., F. Girard-Ardhuin and D. Croizé-Fillon, Sea-ice drift in the Central Arctic using the 89
GHz brightness temperatures of the Advanced Microwave Scanning Radiometer. User’s manual,
Version 2.0, February 2007c.
ftp://ftp.ifremer.fr/ifremer/cersat/products/gridded/psi-drift/documentation/amsr.pdf
Ezraty R. and J. F. Piollé, SeaWinds on QuikSCAT Polar Sea Ice Grids, User Manual.
CONVECTION report 5, V1.1, August 2001. Greenland Sea Convection Mechanism and Their
Implications, Fifth Framework Programme of the European Commission 1998-2002, Contract N°
EVK2-2000-00058.
Gohin, F., A. Cavanié and R. Ezraty, Evolution of passive and active microwave signatures of a large
sea ice feature during its two and half year drift through the Arctic Ocean, J. Geophys. Res., Vol. 103,
C4, pp. 8177-8189, 1998.
Kaleschke, L., C. Lüpkes, T. Vihma, J. Haarpaintner, A. Bochert, J. Hartmann and G. Heygster,
SSM/I Sea ice remote sensing for mesoscale ocean-atmosphere interaction analysis, Canadian Journal
of Remote Sensing, Vol. 27, n° 5, pp. 526-537, 2001
Kwok R., A. Schweiger, D. A. Rothrock, S. Pang and C. Kottmeier, Sea ice motion from satellite
passive microwave imagery assessed with ERS SAR and buoy motions, J. Geophys. Res., Vol. 103,
C4, pp. 8191-8214, April 1998
Liu A.K. and D.J. Cavalieri, On sea ice drift from the wavelet analysis of the Defense Meteorological
Satellite Program (DMSP) Special Sensor Microwave Imager (SSM/I) data. Int. J. Remote Sensing,
Vol. 19, No. 7, pp. 1,415-1,423. 1998
Sea ice drift in the central Arctic combining QuikSCAT & SSM/I sea ice drift data – User’s manual
PolarView Ref. : - C2 – MUT – W – 05 – IF V2.0 22
Liu A.K., Y. Zhao and S.Y. Wu, Artic sea ice drift from wavelet analysis of NSCAT and special
sensor microwave imager data, J. Geophys. Res., Vol. 104, C5, pp. 11,529-11,538, May 1999
Martin T. and E. Augstein, Large scale drift of Arctic sea ice retrieved from passive microwave
satellite data, J. Geophys. Res., Vol. 105, C4, pp. 8775-8788, April 2000.
Sea ice drift in the central Arctic combining QuikSCAT & SSM/I sea ice drift data – User’s manual
PolarView Ref. : - C2 – MUT – W – 05 – IF V2.0 23
Contact
The best source of information is CERSAT
: http://www.ifremer.fr/cersat/
CERSAT - IFREMER
BP 70
29280 Plouzané, France
Phone (33) 2 98-22-46-91
Fax (33) 2 98-22-45-33
For more information on CERSAT archiving and processing facility (FPAF), please
contact :
fpaf@ifremer.fr
M
r
Denis CROIZE-FILLON
Phone (33) 2 98-22-47-12
Fax (33) 2 98-22-45-33
Internet : denis.croize.fillon@ifremer.fr
For more information on PSI grids and products, please contact :
M
rs
Fanny GIRARD-ARDHUIN
Phone (33) 2 98-22-42-99
Fax (33) 2 98-22-45-33
Internet : fanny.ardhuin@ifremer.fr
Sea ice drift in the central Arctic combining QuikSCAT & SSM/I sea ice drift data – User’s manual
PolarView Ref. : - C2 – MUT – W – 05 – IF V2.0 24
ANNEX
1. Unavailable & incomplete QuikSCAT data files
The list is in the QuikSCAT user’s manual at
ftp://ftp.ifremer.fr/ifremer/cersat/products/gridded/psi-drift/documentation/quikscat.pdf
2. Unavailable & incomplete SSM/I data files
The list is in the SSM/I user’s manual at
ftp://ftp.ifremer.fr/ifremer/cersat/products/gridded/psi-drift/documentation/ssmi.pdf
3. Unavailable & incomplete Merged data files
unavailable (m=missing) / partial-coverage (c)
drift files
Merged
drift
maps 3-day 6-day
Winter 1991/1992
1991/12/03 (c)
1991/12/04 (c)
1991/12/03-1991/12/06 (c)
1991/12/04-1991/12/07 (c)
1991/12/03-1991/12/09 (c)
1991/12/04-1991/12/10 (c)
1991/12/24 (c) 1991/12/21-1991/12/24 (c)
1991/12/24-1991/12/27 (c)
1991/12/30 (c)
1991/12/31 (c)
1991/12/27-1991/12/30 (c)
1991/12/28-1991/12/31 (c)
1991/12/30-1992/01/02 (c)
1991/12/31-1992/01/03 (c)
1991/12/18-1991/12/24 (c)
1991/12/24-1991/12/30 (c)
1991/12/25-1991/12/31 (c)
1991/12/30-1992/01/05 (c)
1991/12/31-1992/01/06 (c)
1992/01/08 (c) 1992/01/05-1992/01/08 (c)
1992/01/08-1992/01/11 (c)
1992/01/02-1992/01/08 (c)
1992/01/08-1992/01/14 (c)
1992/01/13 (c) 1992/01/10-1992/01/13 (c)
1992/01/13-1992/01/16 (c)
1992/01/07-1992/01/13 (c)
1992/01/13-1992/01/19 (c)
1992/02/02 (c) 1992/01/30-1992/02/02 (c)
1992/02/02-1992/02/05 (c)
1992/01/27-1992/02/02 (c)
1992/02/02-1992/02/08 (c)
1992/02/14 (c) 1992/02/11-1992/02/14 (c)
1992/02/14-1992/02/17 (c)
1992/02/08-1992/02/14 (c)
1992/02/14-1992/02/20 (c)
1992/04/29 (c)
1992/04/30 (c)
1992/04/26-1992/04/29 (c)
1992/04/27-1992/04/30 (c)
1992/04/29-1992/05/02 (c)
1992/04/30-1992/05/03 (c)
1992/04/23-1992/04/29 (c)
1992/04/24-1992/04/30 (c)
1992/04/29-1992/05/05 (c)
1992/04/30-1992/05/06 (c)
1992/05/27 (c) 1992/05/24-1992/05/27 (c)
1992/05/27-1992/05/30 (c)
1992/05/21-1992/05/27 (c)
1992/05/27-1992/06/02 (c)
Sea ice drift in the central Arctic combining QuikSCAT & SSM/I sea ice drift data – User’s manual
PolarView Ref. : - C2 – MUT – W – 05 – IF V2.0 25
Winter 1992/1993
1992/12/30 (c)
1992/12/31 (c)
1992/12/27-1992/12/30 (c)
1992/12/28-1992/12/31 (c)
1992/12/30-1993/01/02 (c)
1992/12/31-1993/01/03 (c)
1992/12/24-1992/12/30 (c)
1992/12/25-1992/12/31 (c)
1992/12/30-1993/01/05 (c)
1992/12/31-1993/01/06 (c)
1993/01/04 (m) 1993/01/01-1993/01/04 (m)
1993/01/04-1993/01/07 (m)
1992/12/29-1993/01/04 (m)
1993/01/04-1993/01/10 (m)
1993/01/08 (c) 1993/01/05-1993/01/08 (c)
1993/01/08-1993/01/11 (c)
1993/01/02-1993/01/08 (c)
1993/01/08-1993/01/14 (c)
1993/01/17 (c) 1993/01/14-1993/01/17 (c)
1993/01/17-1993/01/20 (c)
1993/01/11-1993/01/17 (c)
1993/01/17-1993/01/23 (c)
1993/02/27 (c) 1993/02/24-1993/02/27 (c)
1993/02/27-1993/03/02 (c)
1993/02/21-1993/02/27 (c)
1993/02/27-1993/03/05 (c)
1993/03/01 (c) 1993/02/26-1993/03/01 (c)
1993/03/01-1993/03/04 (c)
1993/02/23-1993/03/01 (c)
1993/03/01-1993/03/07 (c)
1993/03/11 (c) 1993/03/08-1993/03/11 (c)
1993/03/11-1993/03/14 (c)
1993/03/05-1993/03/11 (c)
1993/03/11-1993/03/17 (c)
Winter 1993/1994
1993/11/17 (c) 1993/11/14-1993/11/17 (c)
1993/11/17-1993/11/20 (c)
1993/11/11-1993/11/17 (c)
1993/11/17-1993/11/23 (c)
1993/12/09 (c)
1993/12/11 (c)
1993/12/06-1993/12/09 (c)
1993/12/08-1993/12/11 (c)
1993/12/09-1993/12/12 (c)
1993/12/11-1993/12/14 (c)
1993/12/16 (c)
until
1993/12/18 (c)
1993/12/13-1993/12/16 (c)
until
1993/12/18-1993/12/21 (c)
1993/12/03-1993/12/09 (c)
1993/12/05-1993/12/11 (c)
1993/12/09-1993/12/15 (c)
until
1993/12/12-1993/12/18 (c)
1993/12/16-1993/12/22 (c)
1993/12/17-1993/12/23 (m)
1993/12/18-1993/12/24 (c)
1993/12/28 (c) 1993/12/25-1993/12/28 (c)
1993/12/28-1993/12/31 (c)
1993/12/22-1993/12/28 (c)
1993/12/28-1994/01/03 (c)
1994/01/01 (c) 1993/12/29-1994/01/01 (c)
1994/01/01-1994/01/04 (c)
1993/12/26-1994/01/01 (c)
1994/01/01-1994/01/07 (c)
1994/02/17 (c) 1994/02/14-1994/02/17 (c)
1994/02/17-1994/02/20 (c)
1994/02/11-1994/02/17 (c)
1994/02/17-1994/02/23 (c)
1994/05/09 (c) 1994/05/06-1994/05/09 (c)
1994/05/09-1994/05/12 (c)
1994/05/03-1994/05/09 (c)
1994/05/09-1994/05/15 (c)
Winter 1994/1995
1994/11/06 (c) 1994/11/03-1994/11/06 (c)
1994/11/06-1994/11/09 (c)
1994/10/31-1994/11/06 (c)
1994/11/06-1994/11/12 (c)
1994/11/20 (m)
1994/11/21 (m)
1994/11/22 (c)
1994/11/17-1994/11/20 (m)
1994/11/18-1994/11/21 (m)
1994/11/19-1994/11/22 (c)
1994/11/20-1994/11/23 (m)
1994/11/21-1994/11/24 (m)
1994/11/22-1994/11/25 (c)
1994/11/14-1994/11/20 (m)
1994/11/15-1994/11/21 (m)
1994/11/18-1994/11/22 (c)
1994/11/20-1994/11/26 (m)
1994/11/21-1994/11/27 (m)
1994/11/22-1994/11/28 (c)
1995/05/03 (c)
1995/05/06 (c)
1995/04/30-1995/05/03 (c)
1995/05/03-1995/05/06 (c)
1995/05/06-1995/05/09 (c)
1995/04/27-1995/05/03 (c)
1995/04/30-1995/05/06 (c)
1995/05/03-1995/05/09 (c)
1995/05/06-1995/05/12 (c)
Sea ice drift in the central Arctic combining QuikSCAT & SSM/I sea ice drift data – User’s manual
PolarView Ref. : - C2 – MUT – W – 05 – IF V2.0 26
Winter 1995/1996
1995/12/19 (c)
1995/12/20 (c)
1995/12/16-1995/12/19 (c)
1995/12/17-1995/12/20 (c)
1995/12/19-1995/12/22 (c)
1995/12/20-1995/12/23 (c)
1995/12/13-1995/12/19 (c)
1995/12/14-1995/12/20 (c)
1995/12/19-1995/12/25 (c)
1995/12/20-1995/12/26 (c)
1996/04/07 (c)
1996/04/08 (c)
1996/04/02-1996/04/07 (c)
1996/04/05-1996/04/08 (c)
1996/04/07-1996/04/10 (c)
1996/04/08-1996/04/11 (c)
1996/04/01-1996/04/07 (c)
1996/04/02-1996/04/08 (c)
1996/04/07-1996/04/13 (c)
1996/04/08-1996/04/14 (c)
1996/04/21 (c) 1996/04/18-1996/04/21 (c)
1996/04/21-1996/04/24 (c)
1996/04/15-1996/04/21 (c)
1996/04/21-1996/04/27 (c)
1996/05/11 (c) 1996/05/08-1996/05/11 (c)
1996/05/11-1996/05/14 (c)
1996/05/05-1996/05/11 (c)
1996/05/11-1996/05/17 (c)
1996/05/30 (c)
1996/05/31 (c)
1996/06/01 (c)
1996/05/27-1996/05/30 (c)
until
1996/05/31-1996/06/03 (c)
1996/05/24-1996/05/30 (c)
1996/05/25-1996/05/31 (c)
1996/05/26-1996/06/01 (c)
1996/05/30-1996/06/05 (c)
1996/05/31-1996/06/07 (c)
Winter 1996/1997
1996/11/02 (c) 1996/10/24-1996/11/02 (c)
1996/11/02-1996/11/05 (c)
1996/10/27-1996/11/02 (c)
1996/11/02-1996/11/08 (c)
1996/11/30 (c)
1996/12/01 (m)
1996/11/27-1996/11/30 (c)
1996/11/28-1996/12/01 (m)
1996/11/30-1996/12/03 (c)
1996/12/01-1996/12/04 (m)
1996/11/24-1996/11/30 (c)
1996/11/25-1996/12/01 (m)
1996/11/30-1996/12/06 (c)
1996/12/01-1996/12/07 (m)
1997/01/31 (c) 1997/01/28-1997/01/31 (c)
1997/01/31-1997/02/03 (c)
1997/01/25-1997/01/31 (c)
1997/01/31-1997/02/06 (c)
1997/02/15 (c) 1997/02/12-1997/02/15 (c)
1997/02/15-1997/02/18 (c)
1997/02/09-1997/02/15 (c)
1997/02/15-1997/02/21 (c)
1997/04/15 (c) 1997/04/12-1997/04/15 (c)
1997/04/15-1997/04/18 (c)
1997/04/09-1997/04/15 (c)
1997/04/15-1997/04/21 (c)
1997/05/31 (c) 1997/05/28-1997/05/31 (c)
1997/05/31-1997/06/03 (c)
1997/05/25-1997/05/31 (c)
1997/05/31-1997/06/06 (c)
Winter 1998/1999
1998/03/10 (c)
1998/03/11 (m)
1998/03/07-1998/03/10 (c)
1998/03/08-1998/03/11 (m)
1998/03/10-1998/03/13 (c)
1998/03/11-1998/03/14 (m)
1998/03/04-1998/03/10 (c)
1998/03/05-1998/03/11 (m)
1998/03/10-1998/03/16 (c)
1998/03/11-1998/03/17 (m)
1999/01/16 (c) 1999/01/13-1999/01/16 (c)
1999/01/16-1999/01/19 (c)
1999/01/10-1999/01/16 (c)
1999/01/16-1999/01/22 (c)
1999/01/26 (c) 1999/01/23-1999/01/26 (c)
1999/01/26-1999/01/29 (c)
1999/01/20-1999/01/26 (c)
1999/01/26-1999/02/01 (c)
Winter 1999/2000
1999/11/08 (c)
1999/11/09 (c)
1999/11/05-1999/11/08 (c)
1999/11/06-1999/11/09 (c)
1999/11/08-1999/11/11 (c)
1999/11/09-1999/11/12 (c)
1999/11/02-1999/11/08 (c)
1999/11/03-1999/11/09 (c)
1999/11/08-1999/11/14 (c)
1999/11/09-1999/11/15 (c)
1999/11/17 (c)
1999/11/18 (c)
1999/11/14-1999/11/17 (c
1999/11/15-1999/11/18 (c)
1999/11/17-1999/11/20 (c)
1999/11/18-1999/11/21 (c)
1999/11/11-1999/11/17 (c)
1999/11/12-1999/11/18 (c)
1999/11/17-1999/11/23 (c)
1999/11/18-1999/11/24 (c)
2000/04/15 (c) 2000/04/12-2000/04/15 (c)
2000/04/15-2000/04/18 (c)
2000/04/09-2000/04/15 (c)
2000/04/15-2000/04/21 (c)
Sea ice drift in the central Arctic combining QuikSCAT & SSM/I sea ice drift data – User’s manual
PolarView Ref. : - C2 – MUT – W – 05 – IF V2.0 27
Winter 2000/2001
2000/12/01 (m)
2000/12/02 (c)
2000/11/28-2000/12/01 (m)
2000/11/29-2000/12/02 (c)
2000/12/01-2000/12/04 (m)
2000/12/02-2000/12/05 (c)
2000/11/25-2000/12/01 (m)
2000/11/26-2000/12/02 (c)
2000/12/01-2000/12/07 (m)
2000/12/02-2000/12/08 (c)
Winter 2006/2007
2006/12/31 (m)
2006/12/28-2006/12/31 (m)
2006/12/31-2007/01/03 (m)
2006/12/25-2006/12/31 (m)
2006/12/31-2007/01/06 (m)
Table 2 : Unavailable and incomplete data as of February 2
nd
2007.
Sea ice drift in the central Arctic combining QuikSCAT & SSM/I sea ice drift data – User’s manual
PolarView Ref. : - C2 – MUT – W – 05 – IF V2.0 28
4. Drift vector density for four winters
4.1 Winter 1999-2000
Oct
1999 Nov
1999 Dec
1999 Jan
2000 Feb
2000 Mar
2000 Apr
2000 May
2000
0.0
0.2
0.4
0.6
0.8
1.0
Drift data density
3-day lag SSMI-H
SSMI-V
QuikSCAT
Merged
Oct
1999 Nov
1999 Dec
1999 Jan
2000 Feb
2000 Mar
2000 Apr
2000 May
2000
0.0
0.2
0.4
0.6
0.8
1.0
Drift data density
6-day lag SSMI-H
SSMI-V
QuikSCAT
Merged
Figure 10 : Time series of drift data density for the winter 1999-2000 for Merged, QuikSCAT
and SSM/I at 3 and 6 day-lags.
Sea ice drift in the central Arctic combining QuikSCAT & SSM/I sea ice drift data – User’s manual
PolarView Ref. : - C2 – MUT – W – 05 – IF V2.0 29
4.2 Winter 2000-2001
Oct
2000 Nov
2000 Dec
2000 Jan
2001 Feb
2001 Mar
2001 Apr
2001 May
2001
0.0
0.2
0.4
0.6
0.8
1.0
Drift data density
3-day lag SSMI-H
SSMI-V
QuikSCAT
Merged
Oct
2000 Nov
2000 Dec
2000 Jan
2001 Feb
2001 Mar
2001 Apr
2001 May
2001
0.0
0.2
0.4
0.6
0.8
1.0
Drift data density
6-day lag SSMI-H
SSMI-V
QuikSCAT
Merged
Figure 11 : Time series of drift data density for the winter 2000-2001 for Merged, QuikSCAT
and SSM/I at 3 and 6 day-lags.
Sea ice drift in the central Arctic combining QuikSCAT & SSM/I sea ice drift data – User’s manual
PolarView Ref. : - C2 – MUT – W – 05 – IF V2.0 30
4.3 Winter 2001-2002
Oct
2001 Nov
2001 Dec
2001 Jan
2002 Feb
2002 Mar
2002 Apr
2002 May
2002
0.0
0.2
0.4
0.6
0.8
1.0
Drift data density
3-day lag SSMI-H
SSMI-V
QuikSCAT
Merged
Oct
2001 Nov
2001 Dec
2001 Jan
2002 Feb
2002 Mar
2002 Apr
2002 May
2002
0.0
0.2
0.4
0.6
0.8
1.0
Drift data density
6-day lag SSMI-H
SSMI-V
QuikSCAT
Merged
Figure 12 : Time series of drift data density for the winter 2001-2002 for Merged, QuikSCAT
and SSM/I at 3 and 6 day-lags.
Sea ice drift in the central Arctic combining QuikSCAT & SSM/I sea ice drift data – User’s manual
PolarView Ref. : - C2 – MUT – W – 05 – IF V2.0 31
4.4 Winter 2002-2003
Oct
2002 Nov
2002 Dec
2002 Jan
2003 Feb
2003 Mar
2003 Apr
2003 May
2003
0.0
0.2
0.4
0.6
0.8
1.0
Drift data density
3-day lag SSMI-H
SSMI-V
QuikSCAT
Merged
Oct
2002 Nov
2002 Dec
2002 Jan
2003 Feb
2003 Mar
2003 Apr
2003 May
2003
0.0
0.2
0.4
0.6
0.8
1.0
Drift data density
6-day lag SSMI-H
SSMI-V
QuikSCAT
Merged
Figure 13 : Time series of drift data density for the winter 2002-2003 for Merged, QuikSCAT
and SSM/I at 3 and 6 day-lags.
... Sea ice drift has been used extensively for the validation of sea ice rheologies in the Sea Ice Model Intercomparison Project (SIMIP; Lemke et al., 1997; Kreyscher et al., 2000). Here we compare sea ice drift from two satellite products (Fowler, 2003 and Ezraty and Piollé, 2004) and buoy data (Ortmeyer and Rigor, 2004) with a number of AOMIP model results for the period of satellite coverage. The observations are presented in the next section. ...
... A second satellite derived drift product is obtained from the Centre ERS d'Archivage et de Traitement (CERSAT). Here, we choose a merged product of Quick Scatterometer (QuikSCAT) and SSM/I derived sea ice drift vector fields (Ezraty and Piollé, 2004), which are projected on a grid that is oriented exactly as NSIDC's SSM/I-12.5km-grid but with a spatial resolution of 62.5 km and covering the central Arctic only. ...
... The ice drift data set from IFREMER, Brest, France used here combines SSM/I and SSMIS microwave radiometer data with QuikSCAT and ASCAT microwave scatterometer data (Ezraty et al., 2007;Girard-Ardhuin et al., 2008) (ftp://ftp.ifremer.fr/ifremer/cersat/products/gridded/psi-drift/data/arctic/ merged-quikscat-ssmi/). ...
Article
Full-text available
The Fram Strait sea ice volume export 1992–2014 is derived by combining sea ice thickness from upward looking sonars (ULS) with satellite observations of sea ice drift and area. Fram Strait is the main gate for sea ice export from the Arctic. The average yearly sea ice export is 2,400 ± 640 km³. The mean and modal ULS ice thickness in Fram Strait decreased by 15% and 21% per decade, respectively, during 1990–2014. Combined with sea ice drift and area this leads to a decrease of the Arctic sea ice volume export of 27 ± 2% per decade between 1992 and 2014. Thus, for the given time period, changes in sea ice export do not drive the sea ice volume decrease in the Arctic Basin. However, for individual years like 2007 and 2012 the ice export likely has contributed to the loss of summer sea ice. Combined with PIOMAS model simulation we estimate that 14% of the total Arctic sea ice volume is exported every year through Fram Strait. This fraction of the total sea ice volume exported per year does not show a trend because the Arctic Basin ice volume is decreasing at a similar rate as the Fram Strait ice volume export. Compared to ice velocities from Acoustic Doppler Current Profiler (ADCP) the satellite ice drift shows good correspondence in variability but a negative bias. Ice volume transport estimates presented here thus should be considered a conservative estimate. We show, however, that the transport estimates are not sensitive to the exact flux gate location.
... The synthesizing process is also completed at IFREMER. Through the combination, the synergy is expected to have an improved data density compared with any single products (Ezraty et al. 2007a). A detailed description related to the original data processing and product assessment is given by Ezraty et al. (2007). ...
Article
Full-text available
Sea-ice outflow from the Laptev Sea is of considerable importance in maintaining the Arctic Ocean sea-ice budget. In this study, a method exclusively using multiple satellite observations is used to calculate sea-ice volume flux across the eastern boundary (EB) and northern boundary (NB) of the Laptev Sea during the October–November and February–March or March–April periods (corresponding to the ICESat autumn and winter campaigns) between 2003 and 2008. Seasonally, the mean total ice volume flux (i.e., NB+EB) over the investigated autumn period (1.96 km3/day) is less than that over the winter period (2.57 km3/day). On the other hand, the large standard deviations of the total volume flux, 3.45 and 0.91 km3/day for the autumn and winter campaigns, indicate significant interannual fluctuations in the calculated quantities. A statistically significant (P>0.99) positive correlation, R=0.88 (or 0.81), is obtained between volume flux across the EB (or NB) and mean ice-drift speed over the boundary for the considered 11 ICESat campaigns. In addition, statistics show that a large fraction of the variability in volume flux across the NB over the 11 investigated campaigns, roughly 40%, is likely explained by ice thickness variability. On average, flux through the Laptev Sea amounts to approximately one-third of that across Fram Strait during the autumn and winter campaigns. These large contributions of sea ice from the Laptev Sea demonstrate its importance as an ice source, affecting the entire sea-ice mass balance in the Arctic Ocean.
... The description given above applies to the well known MCC (Maximum Cross Correlation) technique which has been successfully applied by many investigators (Emery et al. (1991); Ezraty et al. (2008); Haarpaintner (2006); Notarstefano et al. (2007); Schmetz et al. (1993), among others). In the MCC, the search for the best candidate block is discrete and exhaustive. ...
Technical Report
Full-text available
ATBD for the newest version of the OSISAF Low Resolution Sea Ice Drift product, including summer ice drift and uncertainties. More details and data access from: http://osisaf.met.no
... In order to evaluate the possible impact of scattering and dissipation, we have nested a curvilinear grid configuration of the WAVEWATCH III model into the global wave hindcast of Rascle and Ardhuin [2013]. The polar grid is the same as that used for the daily 12.5 km resolution maps of sea ice concentration derived from the Special Sensor Microwave Imager (SSM/I) [Ezraty et al., 2007]. The SSM/I ice concentration c, and CFSR wind fields were used to force the model. ...
Article
Full-text available
The poorly understood attenuation of surface waves in sea ice is generally attributed to the combination of scattering and dissipation. Scattering and dissipation have very different effects on the directional and temporal distribution of wave energy, making it possible to better understand their relative importance by analysis of swell directional spreading and arrival times. Here we compare results of a spectral wave model – using adjustable scattering and dissipation attenuation formulations – with wave measurements far inside the ice pack. In this case, scattering plays a negligible role in the attenuation of long swells. Specifically, scattering-dominated attenuation would produce directional wave spectra much broader than the ones recorded, and swell events arriving later and lasting much longer than observed. Details of the dissipation process remain uncertain. Average dissipation rates are consistent with creep effects but are 12 times those expected for a laminar boundary layer under a smooth solid ice plate.
... The two main patterns of the Arctic ice current system, namely the Beaufort gyre and the transpolar ice drift are present ( Fig. 2.3, bottom left panel) but they are weaker than in data produced by the CERSAT from QuikSCAT and SSM/I drift vectors (Ezraty and Piollé, 2004). For the comparison, we interpolate monthly drift vectors from a 62.5 km polar stereographic grid to our simulation grid and we average data over all winter seasons (October to April) from 1992 to 2001 ( Fig. 2.3, bottom right panel). ...
Thesis
Full-text available
Current sea ice models are far from simulating the complex behavior of the Arctic ice pack. When used at high resolution, they do not provide realistic deformation fields that are responsible for sea ice ridging and leads opening. The impacts of using a better representation of sea ice dynamics may then be crucial for many applications related to sea ice and in particular for the accuracy of climate change predictions, not only in the Arctic but also at a much larger scale. In this thesis, we adopted a twofold strategy. On the one hand, we started from a widely used sea ice model and we improved the accuracy and performances of the numerical methods used to solve sea ice dynamics. On the other hand, we adopted a completely new approach with the implementation of the elasto-brittle rheology that produces more realistic results. We adapted it for a better integration in classical sea ice models and we confirmed its ability to reproduce the complex behavior of the Arctic ice pack.
... A persistent feature of the model is that, compared with the observations of Bourke and Garrett (1987), it tends to somewhat overestimate the ice thickness along the Siberian coast: by about 1 m in the Kara Sea, 1.5 m in the Laptev Sea, and 2 m in the East Siberian Sea. The two main patterns of the Arctic ice current system, namely the Beaufort gyre and the transpolar ice drift are present (Fig. 3, bottom left panel) but they are weaker than in data produced by the CERSAT from QuikSCAT and SSM/I drift vectors (Ezraty and Piollé, 2004). For the comparison, we interpolate monthly drift vectors from a 62.5 km polar stereographic grid to our simulation grid and we average data over all winter seasons (October to April) from 1992 to 2001 (Fig. 3, bottom right panel). ...
Article
Full-text available
In the Arctic, global warming is particularly pronounced so that we need to monitor its development continuously. On the other hand, the vast and hostile conditions make in situ observation difficult, so that available satellite observations should be exploited in the best possible way to extract geophysical information. Here, we give a résumé of the sea ice remote sensing efforts of the EU project DAMOCLES (Developing Arctic Modeling and Observing Capabilities for Long-term Environmental Studies). The monthly variation of the microwave emissivity of first-year and multiyear sea ice has been derived for the frequencies of the microwave imagers like AMSR-E and sounding frequencies of AMSU, and has been used to develop an optimal estimation method to retrieve sea ice and atmospheric parameters simultaneously. A sea ice microwave emissivity model has been used together with a thermodynamic model to establish relations between the emisivities at 6 GHz and 50 GHz. At the latter frequency, the emissivity is needed for assimilation into atmospheric circulation models, but more difficult to observe directly. A method to determine the effective size of the snow grains from observations in the visible range (MODIS) is developed and applied. The bidirectional reflectivity distribution function (BRDF) of snow, which is an essential input parameter to the retrieval, has been measured in situ on Svalbard during the DAMOCLES campaign, and a BRDF model assuming aspherical particles is developed. Sea ice drift and deformation is derived from satellite observations with the scatterometer ASCAT (62.5 km grid spacing), with visible AVHRR observations (20 km), with the synthetic aperture radar sensor ASAR (10 km), and a multi-sensor product (62.5 km) with improved angular resolution (Continuous Maximum Cross Correlation, CMCC method) is presented. CMCC is also used to derive the sea ice deformation, important for formation of sea ice leads (diverging deformation) and pressure ridges (converging). The indirect determination of sea ice thickness from altimeter freeboard data requires knowledge of the ice density and snow load on sea ice. The relation between freeboard and ice thickness is investigated based on the airborne Sever expeditions conducted between 1928 and 1993.
Article
Empirical error functions for 6 different low-resolution Arctic sea-ice drift products are presented on monthly time-scales. To assess the error statistics of the Eulerian ice-drift products, we use high-resolution Lagrangian sea-ice drift obtained from synthetic aperture radar (SAR) images. We processed the Lagrangian drift to Eulerian drift vectors and used them as a reference for the error assessment. Unlike sea-ice buoy trajectory data traditionally used for that purpose, SAR offers a much larger number of data, which enables us to do a thorough assessment of the error statistics of the Eulerian products under different ice conditions. We find that the error statistics differ between the products and between the seasons. For some products the error is dependent on ice drift speed, while for others the error is rather dependent on ice concentration or on both. The summer ice drifts have roughly a two times larger error than the winter drifts, and show significant mean biases. The calculated empirical error functions allow us to derive uncertainty maps for the respective products. These maps can be used for model validation and data assimilation. This article is protected by copyright. All rights reserved.
Article
Full-text available
An attempt to quantify the temporal variability in the volume composition of Arctic sea ice is presented. Categories of sea ice in the Transpolar Drift in Fram Strait are derived from monthly ice thickness distributions obtained by moored sonars (1990-2011). The inflection points on each side of the old ice modal peak are used to separate modal ice from ice which is thinner and thicker than ice in the modal range. The volume composition is then quantified through the relative amount of ice belonging to each of the three categories thin, modal and thick ice in the monthly ice thickness distributions. The trend of thin ice was estimated to be negative at -9.2% per decade (relative to the long-term mean), which was compensated for by increasing trends in modal and thick ice of 8.1% and 4.9% per decade, respectively. A 7-8 year cycle is apparent in the thin and thick ice records, which may explain a loss of deformed ice since 2007. We also quantify how the categories contribute to the mean ice thickness over time. Thick (predominantly deformed) ice dominates the mean ice thickness, constituting on average 66% of the total mean. Following the loss of deformed ice since 2007, the contribution of thick ice to the mean decreased from 75% to 52% at the end of the record. Thin deformed ice did not contribute to this reduction; it was pressure ridges thicker than 5 m that were lost and hence caused the decrease in mean ice thickness.
Article
Full-text available
Two algorithms have been used in a hybrid scheme in order to obtain sea ice concentration maps at 12 km resolution from Special Sensor Microwave Imager (SSM/I) data. One algorithm is based on the 85 GHz polarization difference. The second algorithm is the NASA Team algorithm using the 19 and 37 GHz SSM/I channels. Using 85 GHz SSM/I data allows a significant resolution improvement compared to 19 GHz SSM/I data. This, however, requires careful consideration of the larger atmospheric opacity affecting SSM/I measurements at 85 GHz (weather influence) which can lead to incorrect sea ice concentration estimates - particularly over open water. Our scheme combines the high spatial resolution achieved with the 85 GHz channels with the almost weather-independently NTA-based decision whether the data points belong to the ice-free ocean or to the ice-covered area. Reference 85 GHz brightness temperatures (tie points) have been estimated using a statistical linear regression method with the aid of independent sea ice concentration reference data. These were derived from aircraft dual-polarized passive microwave measurements at 19 and 37 GHz and optical line scanner images. The evaluation of 30 days of SSM/I data comparing the NTA and 85 GHz ice concentrations indicates a good agreement and a strong linear relationship. ERS-2 SAR and SSM/I data were used to analyze the evolution of the Storfjorden Polynya and the MIZ in the Fram Strait. Two different numerical atmospheric models were used to analyze the effect of a resolution improvement from 50 to 12 km of the sea ice concentration data prescribed to model the atmospheric boundary layer (ABL). It was found that the representation of the MZ substantially influences the modelled ABL temperatures. Temperature profiles obtained with the model using the high resolution sea ice concentration data agree significantly better with the profiles measured by the aircraft.
Article
Although estimation of the total ice concentration from special sensor microwave imagers (SSM/I) has proven to be successful, none of the various algorithms developed to discriminate new and older ice provide satisfying results. While the strong contrast between the emissivity of sea ice and that of open water can be utilized to provide reliable estimators of the total ice concentration, passive microwave characteristics of second-year and multiyear ice may locally evolve in different ways, even during the cold season. Scatterometers, as the active microwave instrument in wind mode (AMI-wind) on board the European Remote Sensing Satellites (ERS), provide backscatter data which have a higher sensitivity to the surface topography of ice and a better stability in time, at a resolution compatible with the SSM/I measurements. Here we present the evolutions of the microwave properties of an ice feature appearing along the shores of Novosibirskiye Ostrova (New Siberian Islands) at the end of July 1992 as the ice ages during its 3-year drift toward the Fram Strait. The track of this well-defined ice surface is easily followed on the maps of the backscatter coefficient provided by the AMI-wind during the cold season. In summer, because of melting, the ice undergoes critical changes which alter its microwave signatures and hamper automatic tracking. Moreover, on approaching the Fram Strait the resolution of the scatterometer is not sufficient to capture the complex and rapid transformations of the ice cover. To compensate for this, buoy data obtained from the International Arctic Buoy Program are used, alone during summers or together with satellite data, to build basin-wide ice displacement fields. These displacement fields, successively applied to each pixel of the ice feature selected, provide a series of Lagrangian observations. During the drift, which ends in May 1995, the active and passive signatures evolve coherently, except for the cold season 1992-1993 when unrealistic multiyear ice concentrations are deduced from the brightness temperatures, which, at that time, are much less stable than the backscatter coefficient over the ice surface tracked, identified as second-year ice.
Article
Observing the motion of sea ice from space is analogous to observing wind stress over the wet oceans; both provide surface forcing for modeling ocean dynamics. Ice motion also directly provides the advective component of the equations governing the mass balance of the sea ice cover. Thus its routine observation from space would be of great value to understanding ice and ocean behavior. To demonstrate the feasibility of creating a global multidecadal ice motion record from satellite passive microwave imagery and to quantitatively assess the errors in the estimated ice motions, we have tracked ice every 3 days in the Arctic Ocean and daily in the Fram Strait and Baffin Bay during the 8 winter months from October 1992 to May 1993 and daily in the Weddell Sea during the 8 winter months from March to October 1992. The method, which has been well used previously, involves finding the spatial offset that maximizes the cross correlation of the brightness temperature fields over 100-km patches in two images separated in time by from 1 to 3 days. The resulting ice motions are compared with contemporaneous buoy- and SAR-derived ice motions. The uncertainties in the displacement vectors, between 5 and 12 km, are better than the spatial resolution of the data. Both 85-GHz data with 12-km spatial resolution and 37-GHz data with 25-km resolution are tracked. These trials with the 37-GHz data are new and show quite surprisingly that the error is only about 1 km larger with these data than with the 12-km 85-GHz data. Errors are typically larger than average in areas of lower ice concentration; in the most dynamic regions, particularly near the ice edge in the Barents and Greenland Seas; and in zones of high shear. These passive microwave ice motions show a large increase in spatial detail over motion fields optimally interpolated from buoy and wind observations, especially where buoy data are virtually absent such as near coasts and in some passages between the Arctic Ocean and its peripheral seas. The feasibility of obtaining ice motion from the 37-GHz data in addition to the 85-GHz data should allow an important record of ice motion to be established for the duration of the scanning multichannel microwave radiometer (SMMR), special sensor microwave/imager (SSM/I), and future microwave sensors, that is, from 1978 into the next millenium.
Conference Paper
On an individual orbit basis, QSCAT data does not permit an efficient separation of sea ice from open water. It is shown how the use of the simultaneous H-pol temperatures provided by the radiometer-like processing of the data, QRAD, improves the sorting of sea ice areas from open ocean zones
Article
Based on a nearly linear relationship between Ku-band backscatter and its derivative with respect to incidence angle over Arctic ocean sea ice, maps of backscatter have been produced, using the NSIDC 12.5-km pixel polar stereographic projection grid. Both three-day and one-day maps are studied. The noise level on these maps, evaluated from the difference between successive maps, varies from 2 to 7%, increasing as backscatter level or number of measurements decrease. Spectral analysis indicates the resolution of these maps to be around 40 km, that of the 25-km pixel maps around 60 km. Comparison with RADARSAT Widescan mode scenes, north of Spitsbergen and around Novaya Zemlya, confirm estimations of ice edge and help in interpretation of the maps, while indicating limitations of the water/ice discrimination algorithm
Dynamique de la banquise de l'Océan Glacial Arctique. Rapport d'avancement des travaux
  • F Girard-Ardhuin
Girard-Ardhuin, F., Dynamique de la banquise de l'Océan Glacial Arctique. Rapport d'avancement des travaux. (in French), Rapport Scientifique DOPS/LOS -CNES/IFREMER, Avril 2005.
Validation, critique et fusion de champs de dérive des glaces de mer en Arctique. Rapport de projet de fin d'étude
  • C Carlut
Carlut, C., Validation, critique et fusion de champs de dérive des glaces de mer en Arctique. Rapport de projet de fin d'étude. Rapport IFREMER/DOPS/LOS 2003/03 (in French), 2003
Ardhuin and D. Croizé-Fillon, Sea-ice drift in the Central Arctic using the 89
  • R Ezraty
  • F Girard
Ezraty R., F. Girard-Ardhuin and D. Croizé-Fillon, Sea-ice drift in the Central Arctic using the 89