Retrieval of Thin Ice Thickness from Multi

Sea Ice Climate Change Initiative: Phase 1
Consortium Members
ESA UNCLASSIFIED - For Official Use
Change Record
Issue
Date
Reason for Change
Author
1.0
6/6/2012
First Issue
P. Wadhams
1.1
20/8/2012
Comments included from S. Kern, P Wadhams & V. Djepa
Kjell Kloster (NERSC)
Authorship
Role
Name
Signature
Written by: Prof. P. Wadhams / V. Djepa
Checked by: G. Timms, S. Kern, K. Kloster
Approved by:
Authorised by:
Distribution
Organisation
Names
Contact Details
ESA
Pascal Lecomte
[email protected]
NERSC
Stein Sandven, Lasse H.
Pettersson, Natalia Ivanova,
Johnny A. Johannessen
[email protected];
[email protected]
[email protected]
[email protected]
Logica
Gary Timms, Ed Pechorro, Rehan [email protected];
Raja
[email protected];
[email protected]
Met.no
DMI
DTU
Thomas Lavergne, Lars Anders
Breivik
[email protected];
Leif Toudal Pedersen, Rasmus
Tonboe
[email protected];
Roberto Saldo, René Forsberg,
Henning Skriver, Henriette
Skourup
[email protected];
[email protected]
[email protected]
[email protected];
[email protected]
[email protected]
FMI
Marko Mäkynen, Eero Rinne, Ari [email protected];
Seina
[email protected];
[email protected]
UCL
New point of contact needed
2
University of Hamburg
Stefan Kern
[email protected]
University of Bremen
Georg Heygster
[email protected]
University of Cambridge Peter Wadhams, Vera Djepa,
John Fletcher
[email protected],
[email protected],
[email protected]
MPI-M
Dirk Notz
[email protected]
Ifremer
Fanny Ardhuin
[email protected]
3
Table of Contents
1 Introduction ................................................................................................. 8
1.1 Purpose................................................................................................... 8
1.2 Scope ....................................................................................................8
1.3 Document Structure .................................................................................. 8
1.4 Document Status ...................................................................................... 8
1.5 Applicable Documents ............................................................................... 9
1.6 Reference Documents................................................................................ 9
1.7 Acronyms and Abbreviations ..................................................................... 15
2 Error characterization for sea ice concentration ......................................... 16
2.1 Definition of terms ................................................................................... 16
2.2 Error sources .......................................................................................... 16
2.2.1.1 Error sources for SIC retrieval using satellite observations in the VIS/IR
spectral range. ..................................................................... 17
2.2.1.2 Sources of errors for SIC retrieval using satellite observations in the
microwave spectral range ...................................................... 17
2.2.1.3 Uncertainties of retrieved ice concentration from passive microwave
instruments ......................................................................... 17
2.2.1.4 Uncertainties of the retrieved SIC from active microwave SAR
observations ........................................................................ 19
2.3 Methodology for determination of uncertainties ............................................ 20
2.4 Errors from merging................................................................................. 21
2.5 Effects of smearing, footprint mismatch, tie-point, etc. ................................. 21
3 Error characterisation for sea ice thickness ................................................ 23
3.1 Overview of the error ............................................................................... 23
3.2 Errors associated with the determination of sea ice freeboard ........................ 25
3.3 Errors associated with the conversion of the freeboard derived from RA to SIT . 32
3.4 Uncertainty of the retrieved ice thickness from RA ........................................ 36
4 Sea ice thickness, derived from alternative techniques and correponding
uncertainties................................................................................. 38
4.1 Laser altimetry ........................................................................................ 38
4.2 Error sources of SIT retrieved from satellite IR temperature measurements ..... 38
4.3 Salinity dependence of sea ice thicknes.......................................................38
5 Conclusions……………………………………………………………………………..…………40
4
List of Figures
Figure 2.1. Error analyses of the uncertainties of the derived SIC, using different
algorithms [R35]. ……………………………...............................................................18
Figure 2.2. Comparison of four months of open water fraction estimates from the
Bootstrap and Team algorithms with ice motion derived openings computed within the
RADARSAT RGPS domain...................................................................................20
Figure 2.3. Total uncertainty (standard deviation) of the SSMI product of
OSI/SAF...........................................................................................................21
Figure 2.4. Total, algorithm and smear uncertainty as function of ice
concentration..............................................................................................22
Figure 3.1. Schematic diagram showing the interrelation of the freeboard, sea surface
and ice height, hi [R19]..................................................................................23
Figure 3.2. Break down of the errors associated with a spatially averaged estimate of
sea ice thickness from satellite radar altimetry. Adapted from [Wingham et al., 2001].
................................................................................................................ ......24
Figure 3.3. (a) Histogram representation (left) of the differences of ATM 2 mean
elevations and RA-2 floe type elevations and corresponding scatter plot. (b) Histogram
of ATM 2 mean and RA-2 lead type elevation differences with the corresponding scatter
plot according [R43]..........................................................................................27
Figure 3.4. Histogram representation (left) of the differences of ATM 200 m lead
elevations and RA-2 lead type elevations. ATM 200 m lead elevations are determined
from ATM lead detection analysis. A scatter plot of the ATM and RA-2 lead type
elevations is shown to the right of the floe histogram.............................................28
Figure 3.5. Scatter plots of ASIRAS versus ALS freeboard heights from Fram Strait a)
2008 and b) 2011…………………………………………………………………………………………………………..28
Figure 3.6. Distribution of the 2m air temperature from JRA25 Re-Analysis data for April
24, 2008 (a) and May 3, 2011 (b), 12 UTC. The inset shows the time series of the mean
temperature of the two black boxes for the past 30 days before the experiment took
place, which is indicated by the red arrows............................................................29
Figure 3.7. Scatter plots of RA-2 sea ice freeboard vs. ALS total (sea ice plus snow)
freeboard from Fram Strait 2008 (left) and 2011 (right)…………………………………………..…30
Figure 3.8. Uncertainties of the snow free- board from OIB; a) April, 2009; b) MarchApril, 2010……………………………………………………………………………………………………………………..30
Figure 3.9. Typical ERS altimeter waveforms acquired from (a) a sea ice–covered sea
(specular waveform) and (b) open ocean (diffuse waveform)..................................31
Figure 3.10. from [Eicken et al., 1995]. “Mean density profile and standard deviation of
sea ice density................................................................................................ ..33
5
Figure 3.11. [Alekandrov et al., 2010]. A composite of sea ice density data measurement
ranges obtained from Timco and Frederking, 1996]...................................................34
Figure 3.12. Snow density variations [R26. R19]...................................................35
6
List of Tables
Table 1-1: Applicable Documents.................................................................. 9
Table 1-2: Applicable Standards ................................................................... 9
Table 1-3: Reference Documents ................................................................ 14
Table 1-4: Data sources for validation ........................................................ 12
Table 1-5: Acronyms .....………………………………………………………………………15
7
1
Introduction
A wide range of sea ice surface, underwater, airborne and satellite measurements of sea ice
properties and distribution have been performed in the last decade to retrieve and validate sea ice
concentration and sea ice thickness and corresponding errors.
Satellite microwave (passive (AMSUA/NOAA, SMMR, SSMI, AMSR-E) and active (ASAR/ENVISAT,
SAR/ERS1/2, SAR/RADARSAT-1 and QuikSCAT/SeaWinds) observations proved to be valuable tool for
monitoring sea ice concentration and coverage with some uncertainties due to seasonally varying
physical properties of the sea ice and snow cover. Remote sensing data from laser altimeter on
board ICESat (Ice, Cloud, and land Elevation Satellite), ENVISAT/Radar Altimeter (RA-2) and ERS
altimeter have been used to estimate snow and sea ice freeboard and ice thickness with some
uncertainties due to model parameterization and unknown variables (e.g. snow depth and ice
density).
1.1
Purpose
The purpose of this document is to describe sources of errors and uncertainties of the derived sea
ice properties from remote sensing and surface observations for different retrieval algorithms [R09].
Error characterisation techniques, available data and corresponding retrieval algorithms of sea ice
concentration (SIC) and sea ice thickness (SIT) will be discussed in particular.
1.2
Scope
Overview of the main approaches and algorithms for retrieval of ice concentration
and thickness with corresponding errors will be provided.
Error analyses and uncertainties of the retrieved ice concentration from passive
microwave satellite observations (SSMI, AMSR and SMMR) and optical/infrared
sensors will be provided for selected algorithms (OSI/SAF, NASA) with climate
applications. Current state of sea ice thickness retrieval techniques and sources of
errors will be compared for ice thickness algorithms for laser, ERS and
ENVISAT/RA2 altimeter.
1.3
Document Structure
Error characterisation of sea ice concentration is provided in Section 2.
Error characterisation of sea ice thickness is provided in Section 3.
1.4
Document Status
This is a second revision, incorporating internal and peer review comments.
8
1.5
Applicable Documents
Acronym
Title
Reference
Issue
CSPH
CryoSat Product Handbook
https://earth.esa.int/c/document_li 2012
brary/get_file?folderId=125272&na
me=DLFE-1702.pdf
PMP
Sea Ice ECV project
Management Plan
ESA-CCI_SICCI_PMP_D6.1_v1.1
1.1
PVP
Sea Ice CCI Product
Validation Plan (D2.1)
Error! Reference source not
found.
1.1
DARD
Sea Ice CCI Data Access
SICCI-DARD-04-12
Requirement Document Plan
(D1.3)
1.1
Table 1-1: Applicable Documents
1.6
Reference Documents
Acronym
Title
Reference
Issue
R01
Report on algorithms development http://saf.met.no/docs/rep_algo. 2000
pdf, OSI
R02
AMSR-E Algorithm. Theoretical
Bases Document. Sea ice product
http://wwwghcc.msfc.nasa.gov/A 2004
MSR/atbd/seaiceatbd.pdf,
Markus, T., D. Cavalieri, NASA
R03
http://nsidc.org/data/docs/daac/ 2012
AMSR-E/Aqua Daily L3 25 km
Brightness Temperature & Sea Ice ae_si25_25km_tb_and_sea_ice.g
d.html#markus_cavalieri_2000U
Concentration Polar Grids
CL
R04
CryoSat: A mission to determine
the fluctuations in Earth’s land,
marine ice field
Wingham, D. J., Adv. Space
2006
Res., 37, 841–871,
http://www.cpom.org/research/d
jw-asr37.pdf
R05
CryoSat Data processing concept
http://esamultimedia.esa.int/doc 2001
s/Cryosat/Data_Proc_Concept.pd
f
R06
CryoSat Calibration & Validation
Concept
S. Laxon & T. Pedersen, UCL
R07
Salinity variations in sea ice
Cox, G. N., and W. F. Weeks, J. 1983
Glaciol., 29(12), 306-316
R08
Evaluation of AMSUA and AMSUB
sea ice algorithms for use in the
SAF for OSI
http://www.eumetsat.de/
V.Djepa, SAF
2001
2000
9
Acronym
Title
Reference
Issue
R09
ESA Climate Change initiative
Volume 2 Technical Proposal
www.esa-cci.org,
2011
R10
Validation of the AMSUA and
AMSUB sea ice algorithms for use
in the SAF for Ocean and ice
http://saf.met.no/docs/vsrep_Dj 2000
epa2.pdf
R11
Remote sensing of sea ice
concentration using the Advanced
Microwave Sounder
V.Djepa. RSPSoc, UK.
R12
Satellite remote sensing of ice
V.Djepa, R.Vaughan, RSPSoc
concentration using Synthetic
Aperture Radar data from ERS and
RADARSAT
R13
Global Sea Ice Concentration
http://saf.met.no/p/ice/, Product
Reprocessing, Product User Manual OSI-409
R14
Comparison of the ASI Ice
Concentration Algorithm with
Landsat-7 ETM+ and SAR
Wiebe, H., G. Heygster, and T. 2009
Markus, IEEE Trans. Geosci.
Rem. Sens., 47(9), 3008-3015
R15
The relation between sea ice
thickness and freeboard in the
arctic
Alexandrov, V., et al., The
Cryosphere, 4, 373-380
R16
Observations of ice thickness and Drucker, R., S.Martin, and R.
frazil ice in the St. Lawrence Island Moritz, J. Geophys. Res.,108,
polynya from satellite imagery,
C5, 3149
ULS, and salinity/ temperature
moorings
R17
Joint Polar Satellite system, JPSS
VII RS SIC
http://jointmission.gsfc.nasa.gov 2012
/science/sciencedocuments/0220
12/474-00094_OAD-VIIRS-SICIPSW_RevA_20120127.pdf
R18
The simulated seasonal variability
of the Ku-band radar Altimeter
effective Scattering surface depth
in sea ice
(http://ocean.dmi.dk/remote_se 2011
nsing/alt_sim_thickness_ppf.pdf)
R. Tonboe, S. Andersen, and S.
Gill
R19
Satellite remote sensing of sea ice Kwok, R., J. Glaciol., 56(200),
thickness and kinematics a review 1129-1140
2010
R20
Estimation of sea ice thickness
Kurtz, N., et al., J. Geophys.
distribution through combination of Res., 114, C10007
snow depth and satellite laser
altimeter data
2009
R21
NASA team sea ice algorithm,
Cavalieri
http://nsidc.org/data/docs/daac/ 2008
nasateam/index.html#descriptio
n
R22
Medium Resolution Sea Ice Drift
Product User Manual, OSISAF
http://saf.met.no/docs/osisaf_ss 2012
2_pum_sea-ice-driftmr_v1p2.pdf,
2004
2004
2010
2003
10
Acronym
Title
Reference
Issue
http://osisaf.met.no
Large Decadal Decline of the Arctic
Multiyear Ice Cover
Comiso, J., J. Climate, 25(4),
R24
ENVISAT RA2/MWR Product
Handbook
http://envisat.esa.int/pub/ESA_D 2010
OC/ENVISAT/RA2-MWR/ra2
mwr.ProductHandbook.2_2.pdf
R25
Retracking range, SWH, sigmanaugt and attitude in CryoSat
conventional ocean data
Smith, W., and R. Scharroo,
2011
Ocean surface topography
Science Team Meeting (19-21st
Oct 2011)
R26
Snow depth on Arctic sea ice
Warren, S. G., et al., J. Climate, 1999
12, 1814-1829
R27
Sea Ice Climate change Initiative,
Phase 1
SICCI-PVP -03-12
2012
R28
Sea Ice Downstream services for
Arctic and Antarctic Users and
Stakeholders
Cambridge
2012
R29
Beamwidth effects on sea ice draft Rodrigues, J., Cold Regions
2011
measurements from U.K.
Science and Technology, 65(2),
submarines
160-171
R30
Simulation of the satellite radar
altimeter sea ice Thickness
retrieval Uncertainty
R31
The relation between Arctic sea ice Doble, M. J., et al., J. Geophys. 2011
surface elevation and draft. A case Res., 116, C00E03,
study using coincident AUV sonar doi:10.1029/2011JC007076
and airborne scanning laser
R32
Integrated observations and
modelling of Arctic sea ice and
atmosphere
Heygster, G., et al., Bull.
2009
American Meteorol. Soc., 90(3),
293-297
R33
Relationship between sea ice
freeboard and draft in the Arctic
Basin, and implications for ice
thickness monitoring
Wadhams et al., J. Geophys.
Res., 97(C12)
R34
Arctic sea ice thickness
characteristics in winter 2004 and
2007 from submarine sonar
transects
Wadhams, P., N. Hughes and J. 2011
Rodrigues. J. Geophys. Res.,
116, C00E02
R35
Inter comparison of passive
microwave sea ice concentration
retrievals over the high-
Andersen, S., et al., J.
Geophys. Res., 112, C08004
R23
2012
1176-1193
Tonboe, R., L. Pedersen, and
C. Haas (www.the-cryospherediscuss.net/3/513/2009/)
2009
1992
2007
11
Acronym
Title
Reference
Issue
concentration Arctic sea ice
R36
Atmospheric signatures in sea-ice Oelke C., Int. J. Rem. Sens.
concentration estimates from
18(5), 1113-1136
passive microwaves: modelled and
observed.
1997
R37
Wave dispersion by Antarctic
Wadhams, P., F. Parmiggiani,
2006
pancake ice from SAR images: a
and C. de Carolis, Proc.
method for measuring ice thickness SEASAR Conference, 23-26 Jan.
2006, Frascati, Italy
R38
Satellite-based estimates of sea ice Spreen, G. Ph.D. thesis, Inst. 2008
volume flux: Applications to the
Of Oceanogr., Univ. of
Fram Strait Region
Hamburg, Hamburg, Germany.
(available at
http://www.sub.unihamburg.
De/opus/volltexte/2008/3776/)
R39
Large-scale surveys of snow depth Kurtz, N. T., and S. L. Farrell.
on Arctic sea ice from operation
Geophys. Res. Lett., 38,
IceBridge
L20505.
2011
R40
Arctic sea ice freeboard from
IceBridge acquisitions in 2009:
Estimates and comparisons with
ICESat
Field investigations of Ku-Band radar
penetration into snow cover on Antarctic
sea ice
2012
R41
Kwok, R., et al., J. Geophys.
Res., 117, C02018
Willatt, R., et al., IEEE Trans.
2010
Geos. Rem. Sens., 48(1), 365–
372
R42
Ku band radar penetration into snow cover Willatt, R., et al., Ann. Glaciol.
on Arctic sea ice using airborne data
52(57), 197-205
2011
R43
Comparison of Envisat radar and airborne
laser altimeter measurements over Arctic
sea ice
2009
R44
ICESat measurements of sea ice freeboard Zwally, H., et al., J. Geophys. Res., 113, 2008
and estimates of sea ice thickness in the
C02S15
Weddell Sea
R45
Thin ice thickness from satellite thermal
imagery
R46
A thermodynamic model for estimating sea Wang, X., J. Key, and Y. Liu, J. Geophys. 2010
and lake ice thickness with optical satellite Res., 115, C12035
data
Connor, L., et al., Rem. Sens. Environ.,
113, 563–570
Yu, Y., and D. Rothrock. J. Geoph. Res., 1996
101(C10), 25,753-25,766
12
Acronym
Title
Reference
Issue
R47
A new view of the underside of Arctic sea
ice
Wadhams, P., J. P. Wilkinson, and S. D. 2006
McPhail, Geophys. Res. Lett., 33,
L04501
R48
Experiences from two-years’ through-ice
AUV deployments in the high Arctic
Doble, M. P., et al., IEEE, Autonomous 2008
Underwater Vehicles, Conference,
978-1-4244-2939-4/08
R49
Autosub under Arctic Sea Ice. RRS James
Clark Ross, Report
Wadhams, P., et al., NERC
R50
ICESat observations of seasonal
and interannual variations of
Yi, D., H. J. Zwally, and J. W.
2011
Robbins, Ann. Glaciol., 52(57),
43-51
sea-ice freeboard and estimated
thickness in the Weddell Sea,
2004
Antarctica (2003–2009)
R51
Ice, Cloud, and land Elevation
Satellite (ICESat) over Arctic sea
ice:
Kwok, R., et al., J. Geophys.
Res., 112, C12013
2007
Retrieval of freeboard
R52
ICESat over Arctic sea ice:
Estimation of snow depth and ice
thickness
R53
Estimation of the thin ice thickness Martin, S., et al., J. Geophys.
and heat flux for the Chukchi Sea Res., 109, C10012
Alaskan coast polynya from Special
Sensor Microwave/ Imager data,
1990–2001
2004
R54
Retrieval of Thin Ice Thickness
from Multifrequency Polarimetric
SAR Data
Kwok, R., et al., Rem. Sens.
Environ., 51(3), 361-374
1995
R55
Sea ice thickness retrieval from
SMOS brightness temperatures
during the Arctic freeze-up period
Kaleschke, L., et al., Geophys.
Res. Lett., 39, L05501
2012
R56
The use of operational ice
Agnew T., and S. Howell,
Atmosphere-Ocean, 41(4),
317-331
2003
charts for evaluating passive
microwave ice concentration
R57
R58
Kwok, R., and G. F.
2008
Cunningham, J. Geophys. Res.,
113, C08010
data
A Comparison of Two 85-GHz SSM/I Ice
Kern, S., L. Kaleschke, and D. A. Clausi, 2003
Concentration Algorithms With AVHRR and IEEE Trans. Geosci. Rem. Sens., 41(10),
ERS-2 SAR Imagery
2294-2306
SSM/I sea ice remote sensing for
mesoscale ocean-atmosphere
interaction analysis
Kaleschke, L., et al., Can. J.
Rem. Sens., 27(5), 526–537
2001
13
Acronym
Title
Reference
Issue
R59
A new method for mediumKern, S., Int. J. Rem. Sens.,
resolution sea ice analysis using
25(21), 4555–4582
weather influence corrected
Special Sensor Microwave /Imager
85 GHz data
R60
SAR Measurements of sea ice.
Chapter 3.
Onstott, R. G., and R. A.
2000
Shuchman
http://www.sarusersmanual.com
/ManualPDF/NOAASARManual_C
H03_pg081_116.pdf
R61
Sea ice concentration estimates from
satellite passive microwave
Kwok, R., Geophys. Res. Lett.
29(9), 1311
R62
R63
R64
R65
R66
radiometry and openings from SAR
ice motion
The Impact of Sea Ice Concentration
Parkinson, C., et al., J. Climate, 14,
Accuracies on Climate Model Simulations 2606-2623
with
the GISS GCM
2004
2002
2001
Freeboard, snow depth and sea ice
Markus, T., et al., Ann. Glaciol. 52(57), 2011
roughness in East Antarctica from in situe 242-248
and multiple satellite data
Equations for determining the gas and brine Cox, G.F.N., and W. F. Weeks, J. Glaciol., 1983
volumes in sea-ice samples
29(102), 316-316
Beaven, S., G. Lockhart, S.
Laboratory measurements of radar Gogineni, A. Hosseinmostafa, K.
backscatter from bare and snow Jezek, A. Gow, D. Perovich, A.
covered saline ice sheets
Fung, S. Tjuatja, Int. J. Rem.
Sens., 16(5), 851–876
1995
Kwok,
Sens.
Environ., 1995
1995
Kwok, R.,
R.,etetal.,
al.,Rem.
Rem.
Sens.
Retrieval of Thin Ice Thickness from Environ., 51(3), 361-374
51(3), 361-374
Multi-frequency Polarimetric SAR Data
Estimation of the thin ice thickness Seelye Martin and Robert
R67 and heat flux for the Chukchi Sea Drucker, Ronald Kwok and
2004
Alaskan coast polynya from Special Benjamin Holt. J. Geophys. Res.,
Sensor
Microwave/
Imager 109, C10012,
data,1990–2001.
http://rkwok.jpl.nasa.gov/publica
tions/Martin.2004.pdf
R68 A
thermodynamic
model
for Wang X., Jeffrey R. Key, and Y.
estimating sea and lake ice thickness Liu. J. Geophysical Research. v.
115, C12035, doi:10.1029/
with optical satellite data
2009JC005857, p.1-14.
2010
Table 1-3: Reference Documents
14
Data
Source
Date
OSI/SAF SIC,
ftp://saf.met.no/reprocessed 1978-2009
/ice/conc/v1p3/
Submarine draft
NSIDC
http://nsidc.org/data/g013
60.html
ULS
http://www.whoi.edu/beaufo
rtgyre/index.html
Submarine sonar
statistics of draft.
DAMTP
2004-2007
Ice draft and
elevation
Ice, type, drift,
concentration
DAMTP and Danish Space
Institute
http://saf.met.no/p/ice/
2007
Location
Arctic / Antarctic
Beaufort sea
Beaufort Sea
2012
Table 1-4: Data sources for validation
1.7
Acronyms and Abbreviations
Acronym
Meaning
AATSR
ATSR
AVHRR
ECMWF
EO
FOV
IR
K
MetOp
MODIS
NIR
NPOESS
PDGS
SST
VS
Advanced Along Track Scanning Radiometer
Along-Track Scanning Radiometer
Advanced Very High Resolution Radiometer
European Centre for Medium Range Weather Forecast
Earth Observation
Field Of View
Infrared
Kelvin
Meteorological Operational Satellite
Moderate Resolution Imaging Spectro-radiometer
Near Infrared
National Polar-Orbiting Operational Environmental Satellite System
Payload Data Ground System
Sea Surface temperature
Visible
Table 1-5: Acronyms
15
2
Error characterization for sea ice concentration
Global Climate Model (GCM) simulations [62] show that uncertainties of +7% in the derived
sea ice concentration (SIC) can affect the simulated air temperature by more than 6oC. The
essential impact of the sea ice extend on model forecast and Earth radiative balance requires
review of the existing algorithms for retrieval of SIC with precise analyses of the
corresponding uncertainties.
Error characterization of the retrieved sea ice concentration will be focussed on algorithms,
satellite products and validation data, designed for climate study, using observations in
atmospheric windows in the optical/infrared and microwave spectral range, buoy and in-situ
measurements. The algorithms for deriving ice concentration and corresponding uncertainties
have been analysed in many documents [R01, R03, R11, R12, R21, R17 and R02] and are
summarised by OSI/SAF EUMETSAT project [R13]. The purpose of this section is to point out
the error sources, uncertainties and methodology for characterisation of the errors of the
derived ice concentration from different observations.
2.1
Definition of terms
The definition of terms is provided in Appendix A.
2.2
Error sources
Considering that the sea ice concentration, Ci, is defined as a fraction of sea ice within a
certain area [R13, R57-R58], the total emission of the area will be:
surface = water(1-Ci) +ice Ci
(2.1)
The total uncertainty of the derived SIC can be estimated as a function of algorithm
sensitivity, alg (to surface emissivity, surface and atmospheric conditions, atmos) and uncertainties
related to the instrument and data processing tech (including instrument noise, noise, and
smearing, smear):
Ci = alg(surface,atmos)+technoise, smear)
(2.2)
The uncertainties of the retrieved SIC can increase in heterogeneous areas (especially in the
ice melting period) due to complex contribution of different ice types and snow cover within
the radiometer foot print, considering that the algorithms are validated for tie points of
homogeneous surfaces of FY, MY ice and water. The greatest uncertainties in the retrieved
SIC are usually in the fall winter season when a presence of different sea ice types (MY, FY,
new ice) can be observed in the radiometer foot print area. The sensitivities of the SIC
retrieval algorithms to the atmospheric parameters (wind speed, water vapour, cloud liquid
water (CLW)) and surface temperature were examined for different ice types and
atmospheric conditions using climate and radiative transfer model (e.g. MWMOD) [R08, R11]
and the algorithms using satellite microwave observations with minimum sensitivity to
atmospheric conditions are selected for estimation of SIC [R10, R35]. Although SIC retrieval
using satellite microwave observation has been the backbone for routine sea ice cover
monitoring for the last three decades we will also briefly take a look at uncertainty sources
for SIC retrieval using satellite data from sensors operating in the optical and IR frequency
range.
16
2.2.1.1 Error sources for SIC retrieval using satellite observations in the
VIS/IR spectral range.
The main sources of errors for algorithms to retrieve SIC from satellite observations in the
optical, only during daylight conditions, and infrared, only during freezing conditions, spectral
range are due to impact of the atmosphere, that is mainly the cloud influence, and surface
emissivity, where the input data could be satellite observations in the visible (VIS) and
Thermal Infra-red (TIR) spectral ranges (AVHRR/MetOp, AATSR/ENVISAT, MODIS,
ASTER/Terra) [R37]. As the TIR data are sensitive to clouds and atmospheric liquid water
content, the SIC data are only produced in areas with clear skies. Data in the VIS spectral
range show higher spectral contrast than TIR data where surface melt occurs (during
summer, melting sea ice and open water have almost the same temperature), but are also
sensitive to atmospheric water content. To avoid the impact of melting, TIR data should only
be used under clear sky conditions in the period from October to April/May.
The strong dependence of snow spectral reflectance in the near infra-red (NIR) spectral range
on snow properties, as well as the brightness temperature dependence on snow depth, allows
us to use satellite observations (MODIS, AATSR) in the visible, NIR and TIR spectral range to
retrieve information about ice and snow properties, important for deriving of ice thickness.
For example, bare thin ice can be separated from thick sea ice on satellite TIR images as it
has warmer temperature than the usually snow covered adjacent thick sea ice. Such
techniques and data could be useful to investigate the uncertainty of the SIC retrieval due to
different ice types, here: thin sea ice. The NPOESS algorithm for deriving ice concentration
from satellite observations in the VIS and TIR spectral ranges with corresponding error
analyses is described in [R13, R17, R32].
The advantage of using sensors operating in the VIS/IR spectral range is that these offer a
superior spatial resolution, often better than 1 km. This reduces uncertainties from smearing,
i.e. open water and sea ice can be separated from each other more clearly, and from a mix of
surface types within one grid cell, like open water, thin sea ice and thick sea ice within one
grid cell [R38].
2.2.1.2 Sources of errors for SIC retrieval using satellite observations in the
microwave spectral range
SIC has been retrieved from passive (SMMR, SSM/I) and active (SAR, Scatterometer) sensors
operating in the microwave spectral range between 5 GHz and 90 GHz for about three
decades. There is a fundamental difference between these two types of sensors. Passive
microwave sensors usually retrieve surface information (brightness temperature) as an
integral over a surface footprint area of the order of a few ten kilometres. Active microwave
sensors can retrieve surface information (radar backscatter) with much finer spatial
resolution which is of the order of a few ten meters to 100 meter if we speak of SAR sensors.
Scatterometers like Metop-A ASCAT or Seawinds QuikSCAT offer spatial resolutions that are
similar to passive microwave data.
2.2.1.3 Uncertainties of retrieved ice concentration from passive microwave
instruments
The algorithms and corresponding uncertainties for SIC retrieval from satellite observations in
the microwave spectral range are summarised in OSI/SAF project [R13] and R11, R12, R03].
Basically we have to distinguish between errors caused by the atmosphere which are not
17
changing the surface emissivities such as the atmospheric water content, and errors caused
by the atmosphere and snow and sea ice property changes influencing the surface emissivity
like wind roughening of the open water, snow property (e.g. wetness) changes, or ice type
changes. The inability to distinguish a greater number of surface types can be a significant
source of error. The presence of a third or even fourth surface type additional to just open
water and sea ice in general cannot be accounted for by most of the algorithms. Algorithms
such as the NASA-Team algorithm allow retrieving sea ice concentration and fraction for the
first-year and multiyear sea ice. This does not necessarily mean that such algorithms provide
the most realistic total sea ice concentration, however.
Most algorithms for retrieval of SIC use a set of fixed values ('tiepoints') for emissivity. Such
tie points reflect either 0% sea ice and 100% sea ice, or are more complex in case that more
surface types can be resolved with the SIC algorithm. Tie points change with frequency and
polarization because the typical emissivity changes. These values may also change with
sensor, e.g. from SMMR to SSM/I to AMSR-E although frequencies are perhaps similar at the
first glance. Such fixed tie points usually do not account for seasonal or ice-type related
changes of the surface emissivity. Sensitivity analyses show that variation in sea ice
emissivity of 0.01 over 100% FY ice corresponds to error of 4.5% in derived SIC. The overall
accuracy of the SMMR SIC is estimated to be ±7% (largely due to instrument noise) and the
mean accuracy of the algorithms, used to compute SIC from SSM/I data, such as NASA Team
and Bootstrap are reported to be 1-6 % in winter [R02,R08, R13]. The presence of melt
ponds on sea ice is a serious source of error. Melt ponds are indistinguishable from the open
water in the leads. SIC biases due to melt ponds can be as high as 40% [R56-R59]. In
addition, depending on the algorithm, there are different ways how atmospheric conditions
influence SIC retrievals. The presence of water vapour, clouds, and precipitation result in a
change in the opacity of the atmosphere. Also, wind causes a roughening of the surface of
open water, resulting in changes to the emissivity. This influence is treated in a different way
by the different algorithms and its impact on SIC varies in such a way that even inter-annual
trends are different between different algorithms [R35].
Figure 2.1. Error analyses of the uncertainties of the derived SIC, using different algorithms
[R35].
18
Some algorithms have the atmospheric influence implicitly in the used tie points (e.g. [R57])
while other algorithms use radiative transfer modelling to correct the used brightness
temperature (e.g. [R58]). The majority of the algorithms, however, use so-called weather
filters. These filters depend on the change of the used retrieval parameters, e.g. the
brightness temperature, by atmospheric parameters, e.g. water vapour. The applied weather
filters decrease atmospheric effects but limit detection of low SIC (e.g. less than 12% FY ice
concentration and less than 8% MY ice). Detailed error analyses of the uncertainties of the
derived SIC for different algorithms is given in [R35].
Random instrument noise results in uncertainties in the brightness temperature and in the
derived SIC. This also varies between algorithms. The uncertainty in the derived SIC from
SSM/I due to sensor noise is less than 2.6 % for the Bristol algorithm and between 1.4 %
and 1.7 % for the Bootstrap algorithm [R22, R13]. Studies that compare different algorithms
have found that differences between algorithms are largest in summer, with differences of
more than ±20% and in winter, differences between the different algorithms are usually less
than ±10% [R35].
2.2.1.4 Uncertainties
observations
of the retrieved
SIC from
active microwave
SAR
Using SAR data to obtain SIC is challenging because of the varying nature of the sea ice as
well as the open water. The radar backscatter coefficient of sea ice and open water depends
on frequency, incident angle, polarisation, ice type (MY or FY), roughness, temperature, wind
speed, snow cover and presence of pools of water over snow (ice) (during summer) [R37,
R60]. One cannot apply a simple threshold to discriminate between open water and sea ice.
For example, both open water and the sea ice surface can be either rough (wind-roughened
sea water—pancake ice) or smooth (calm water—new ice). Therefore, it is sometimes difficult
to discriminate roughened open water from pancake or either sea ice types with radar
signature typical to that of rough sea ice, and calm water from thin ice. Multiyear and FY ice
can be discriminated during the winter independent on the wavelength using frequencies in
the C- to Ku-Band but in the spring and summer the discrimination is difficult because of the
elevated liquid water content in the snow and the free water on the surface (melt ponds),
which prevent the microwave penetration.
Higher resolution SAR data have been used to detect sea ice extent and validation of SIC
derived from passive microwave sensors especially in heterogeneous or coastal zone areas
[R35]. The biases between SIC, retrieved from SAR and passive microwave observations are
estimated between −2.9% and 2.6% and the RMS errors reported in [R14] are from 16.9%
to 20.1%. The biases of the derived SIC from SSMI, using 8 algorithms over 100 % ice and
the averaged SAR derived SIC (99.7%) (applying supervised classification) are between 2.1% and 6.9% and depend also on sensor noise [R35]. A supervised classification, using
neural network has been applied for SIC retrieval from ERS2/SAR data [R57] and comparison
with 2 SSMI algorithms for retrieval of SIC. High correlation coefficient between SIC retrieved
from SAR and SSMI is reported for close to 100 % SIC, but very poor correlation is observed
in area of low SIC, which can be explained with different (passive microwave ) algorithm
sensitivities, wave length, weather filters, spatial resolution and SAR classification algorithm.
The comparison of SIC retrieved from passive microwave instruments with that estimated
from RADARSAT [R61] confirms the insensitivity of the passive microwave instruments to
small areas with open water in winter leads. About 3% difference (Figure 2.2.) in the
estimated open water fraction from passive microwave radiometry and RADARSAT
19
Geophysical Processor System (RGPS) (http://earthdata.nasa.gov/data ) have been reported
(e.g. in R61), which is significant and requires review of algorithm sensitivity.
Figure 2.2. Comparison of four months of open water fraction estimates from the Bootstrap
and Team algorithms with ice motion derived openings computed within the RADARSAT RGPS
domain.
It has to be mentioned in this context, however, that considering the higher spatial resolution
of ASAR and RADARSAT, using different polarisations, many sea ice services depend heavily
on SAR data for their ice charts, applying supervised analysis, and using additional data. (e.g.
SAR products delivered from NOAA, Centre for Satellite Applications and research, Alaska
near real time applications AKDEMO (www.star.nesdis.noaa.gov/)
2.3
Methodology for determination of uncertainties
The methodology for error analyses is based on validation and calibration of the retrieved
SIC, using products obtained from different space instruments, air-borne, ship-based
observations, meteorological data and applying statistical analyses [e.g. R13, R22, R35]. The
uncertainty is usually given in terms of a standard deviation and a quantification of the
confidence we have in a certain data point. Following Equations 2.1-2.2 the algorithm
uncertainty depends on accuracy of estimated surface emission and applied weather filter.
The uncertainty of the retrieved surface emission depends on the uncertainties of the
emissivities of the component surfaces (water, βwater, and ice, βice) and the derived SIC:
surface = water(1-Ci)2 +iceCi
(2.3)
The accuracy of the emissivities of the component surfaces (water, ice (MY, FY)) depends on
the a-priory information (model simulation, tie points, surface or airborne observations), used
to calibrate the algorithm.
The weighted contribution and corresponding standard deviations of other sources of errors
(weather filter, instrument noise, smear, etc.) has to be considered when the total
uncertainty of the SIC product is calculated. More information about the methodology will be
given in the final CECR version which is going to be finished after the round robin exercise
has been carried out.
20
2.4
Errors from merging
Validation of SIC, retrieved from passive microwave observations, using SAR, airborne and
ship observations with different footprints may lead to errors from merging of data. Both the
SMMR and the SSM/I instruments have large foot-prints on the ground and representation of
the satellite brightness temperatures or backscatter coefficient and derived products on a
finer, predefined grid results in merging error in addition to the geo-location error and errors
from the impact of sea ice variability over the sampling period and area.
2.5
Effects of smearing, footprint mismatch, tie-point, etc.
SIC data are often given on a finer grid (typically 12.5 or 25 km) than the sensor resolution
(12 to 50 km). This is sometimes called smearing. The error of the estimated SIC from low
resolution microwave radiometer may increase in regions with dynamic surfaces (e.g. from
water, MY and FY ice).
Figure 2.3. Total uncertainty (standard deviation) of the SSMI product of OSI/SAF
The combination of foot-prints of varying size (depending on the frequency) in the
SIC retrieval algorithm results in an additional smearing effect, called foot-print
mismatch error.
21
Figure 2.4. Total,
concentration.
algorithm
and
smear
uncertainty
as
function
of
ice
The impact of smearing on uncertainties of the retrieved ice concentration from passive
microwave satellite observations, using OSI SAF algorithm is shown on Figure 24 [R13].
During winter months, accurate values for tie points are relatively easy to obtain since large
interior areas of the sea ice are at 100% concentration. In summer however the situation is
complicated by the presence of melt ponds. The correct choice of the tie-points is important
for the SIC retrieval as they also include the mean atmospheric influence. The dynamical tie
points may minimize uncertainty due to the climatic trends in the atmosphere and on the ice
surface. The impact of the tie-points on the retrieved SIC is analysed in [R38] and [R13].
22
3
Error characterisation for Sea Ice Thickness
Satellite radar altimeters on board ERS1, 2, EnviSat and CryoSat-2 provide long term SIT
observations on the spatial scale required for climate studies. Error characterisation of the
derived SIT is required for comparison of long term time series of SIT in the Arctic, for
evaluation of sea ice mass balance change, for improved prediction of climate (GCM) and
numerical weather forecast models (NWM).
This section will be focussed on error characterisation of ice thickness and freeboard retrieval
from radar altimeter (ICESat), ENViSAT/radar Altimeter (RA-2), ERS/Altimeter and
approaches for validation of the retrieved freeboard and thickness.
The document describes the contributing factors to the error in a spatially averaged sea ice
thickness estimate from satellite radar altimetry based on: i) CryoSat calibration and
validation document [Wingham et al., 2001];ii) follow up calibration campaigns; iii) collocated
RRDP data base and sensitivity analyses.
At the end of this document, information about deriving sea ice thickness from satellite laser
altimetry, satellite IR temperature and salinity are discussed also. The section is organized as
follows: §1 provides an overview of the error related with SIT retrieval from RA; §2 describes
the errors associated with the determination of the sea ice freeboard including the sampling
error; §3 describes the errors associated with converting a measurement of freeboard to an
estimate of sea ice thickness; §4 summarise the uncertainties of the ice thickness
measurement from satellite radar altimetry and §5 describes other techniques for deriving
sea ice thickness.
3.1. Overview of the error
The estimates of sea ice thickness from RA are based on assumption that the ice floe is in
hydrostatic equilibrium. Figure 3.1 shows a floe, covered by a layer of snow, floating in the
sea and the relations of the freeboard ( hif ) and ice thickness (hi). The objective of satellite
altimetry is to measure the sea ice thickness h, [R27], which is calculated from the freeboard.
Figure 3.1: Schematic diagram showing the interrelation of the freeboard, sea surface and ice
height, hi [Adapted from Kwok 2010].
The satellite radar altimeter (RA) measures the freeboard of the ice (hfi) by subtracting the
elevation of the water (hssh) from the ice elevation and this measurement is converted to an
estimate of sea ice thickness [R19, R20] by:
23
hi = hfi w/ (w - i) + hs s/(w – i)
(3.1)
where information for snow depth (hs) water (w), sea ice (i) and snow density (s) are
required in addition to the freeboard measurement in order to derive the SIT. The error in the
spatially averaged estimate of sea ice thickness contains contributions from the measurement
of the ice freeboard and from the error in the variables in Equation 3.1. Figure 3.2 provides a
more detailed breakdown of the errors and validation methods
Figure 3.2: Break down of the errors associated with a spatially averaged estimate of sea ice
thickness from satellite radar altimetry. Adapted from [Wingham et al., 2001].
Apart of the error due to contribution of sea ice, snow, water densities and snow depth the
freeboard error depends on parameterisations of the radar wave propagation through the
atmosphere, the Earth’s gravity field (or geoid) and the ocean surface dynamic topography
(determined by ocean tides, atmospheric pressure loading, currents, swell). Various methods
of measuring SIT (upward-looking sonar, electromagnetic sounding, or surface point
measurements) have been used to validate SIT retrieved from satellite data. The main errors
of the SIT derived from radar altimeter data are related with: i) sea ice freeboard retrieval
from RA; ii) SIT retrieval from hfi (Equation 3.1.) and absence of a-priori information for snow
and ice type, density and snow depth.
24
3.2. Errors associated with the determination of sea ice freeboard
The elevation of the sea ice or lead is calculated by fitting a model to determine the first
arrival time (τ) of the echo at the satellite position at point Z from the surface. This echo
delay is converted to elevation by
(3.2)
where c is the velocity of light and hobs is the elevation of an ice floe with respect to the
reference ellipsoid.
The error in the first arrival time (
) has two contributing factors: i) due to the fact that the
retrieval assumes the surface locally to be a plane (
) ; ii) due to contribution from the
instrument ( ) to the error in τ.
Sea ice freeboard is calculated by subtracting an ice elevation from some local average of the
ocean elevation:
(3.3)
where
hssh is
the estimate of the surface of the lead,
if the floe is not present, and
and
are corrections that account for the tides and ocean topography at the
location and time of the measurements of the ice or lead. The elevation of water (SSH) is a
sum of contributions from a number of physical processes:
hssh (x, t) = hg(x) +ha(x, t) + hT(x, t) + hd(x, t)
(3.4)
where hg is associated with geoid undulations, ha represents the atmospheric pressure
loading, hT summarizes tidal contributions, and hd accounts for the ocean dynamic
topography associated with geostrophic surface currents and other surface currents caused,
e.g., by eddies. All these terms vary in time and space and contribute to the uncertainty of
the derived sea ice thickness (SIT) when the sea ice, water and snow height are measured
relatively to the level of a reference ellipsoid. The sea ice freeboard, hfi, can be measured also
as a difference between the sea ice surface or ice-snow interface.
Considering above the error (
) of the sea ice freeboard is:
(3.5)
In the following sub-sections we address
(ocean tide and topography model error),
(retrieval error),
(instrument error),
(propagation error),
(satellite position error).
25
The retrieval error (
) depends on surface roughness and penetration.
Surface roughness
The statistics of the surface roughness of sea ice may not be stationary within the area
illuminated by the altimeter (~ 1 km2). If there are many corrugations within this area then
the retuned echo will be sensitive to their average properties, however if the corrugation is
large, or if it has a particular orientation, then the effect on the echo may be complicated and
the elevation may be biased as a result. Hendricks et al (2010) demonstrated that laser
airborne and RA on board of satellite can be statistically biased by the presence of small
patches of open water or ice deformation zone.
Penetration error
It is generally assumed that the dominating scattering surface for the radar is from the
snow/ice interface. Some observations demonstrate variations in the radar penetration depth
over Arctic snow covered sea ice.
The penetration depth of radar signal depends on the snow properties. If sea ice is covered
by dry, cold snow, Beaven et al. [1995] conclude from laboratory experiments, that a Kuband radar signal at normal incidence reflects at the snow-ice interface. In case of wet snow
the radar signal does not penetrate into the snow layer, but reflects from the snow surface
[Hallikainen, 1992]. Internal ice layers and ice lenses in the snow layer, snow grain size and
the presence of frost flowers affects the penetration depth.
Airborne radar altimeter and in situ field measurements, collected during the CryoSat
Validation Experiment (CryoVEx) on May 2006 and 2008 field campaigns have also been used
to investigate the dominant scattering surface over Arctic sea ice. Giles et al. [2007] found
radar penetration to agree well with expected snow depths in Fram Strait. Results of
measurements carried out north of Greenland by Willatt et al. [2011] show that in 2006 only
25% of the dominant radar return originated from closer to the snow-ice interface than to the
snow surface under close to freezing temperatures while in 2008 this fraction increased to
80% during colder conditions (T2006=-4⁰C and T2008=-8⁰C). Hendricks et al. [2010] finds no
penetration of the ASIRAS radar into the snow layer covering the sea ice in the Lincoln Sea
(outside Alert) for either the 2006 or the 2008 data set. Further they find a small penetration
of the radar signal of the sea ice in the Greenland Sea (Fram Strait), but with obtained
depths less than the expected snow depths. Both studies by Willatt et al. [2011] and
Hendricks et al. [2010] are based on data collected in late spring when the snow might not
be dry and cold any more. Ricker et al. [2012] found reflection somewhere between the snow
surface and the snow-ice interface based on ESA CryoVEx 2011 data, which were collected
earlier in the season (mid-April) during cold dry conditions. A similar result was found by
Willatt et al. [2010] for cold Antarctic snow on sea ice. The results from these investigations
suggest penetration depth dependence on temperature because in 2006, when the snow
temperatures were close to freezing, the dominant scattering surface in 25% of the radar
returns appeared closer to the snow/ice interface than the air/snow interface. However, in
2008, when temperatures were lower, the dominant scattering surface appeared closer to the
snow/ice interface than the air/snow interface in 80% of the returns. It is important to note
that radar altimeter estimates of sea ice thickness are only made during winter (OctoberMarch) as when the snow pack warms and begins to melt the signal is difficult to interpret.
Further CryoVEx experiments have been conducted to investigate the radar penetration into
the snow layer (e.g. CryoVEx 2011).
26
Figure 3.3: (a) Histogram representation (left) of the differences of ATM 2 mean elevations
and RA-2 floe type elevations and corresponding scatter plot. (b) Histogram of ATM 2 mean
and RA-2 lead type elevation differences with the corresponding scatter plot according [R43].
The Laser Radar Altimetry (LaRA) airborne field campaign (May 2002) demonstrated that the
laser-measured surface was consistently higher than the radar-measured surface over snow
covered sea ice, consistent with the hypothesis that the radar penetrates into the snow while
the laser measures the snow/air interface. Connor et al, (2009) compare airborne laser
altimetry (ATM) elevations and elevations from the Envisat radar altimeter (RA-2) over sea
ice and show that the radar elevations are lower than the laser elevations (Figure 3.3). They
find a mean difference in elevation of 0.36 m over floe (flat unbroken surface), which is
consistent with the snow depth climatology from Warren et al (1999) and difference of
0.31cm over leads. According to Connor et al, (2009) the surfaces of refrozen leads are very
flat and smooth, producing quasi-specular returns and an associated rapid drop in return
power with increasing angle off-nadir. Radar returns from ice floes are less specular, resulting
in more diffuse returns and a slower drop in power. Elevations over floes are obtained using
an OCOG retracker (Bamber,1994) whilst those over leads are obtained by fitting a Gaussian
function to the return echo, which explains the difference of elevation over leads and floe. In
the both cases the average difference of the sea ice elevation, estimated by RA2 and ATM is
assumed to be due to snow accumulation and supports the hypothesis that laser (ATM)
measures snow surface elevation while the radar (RA-2) measures the elevation of the
underlying ice/snow interface. For refrozen leads with very little snow or no snow Connor et
al, (2009) found near-zero offset between lead elevation estimates from airborne laser
altimetry (ATM) and the Envisat radar altimeter (RA-2) (Figure 3.4).
27
Figure 3. 4. Histogram representation (left) of the differences of ATM 200 m lead elevations
and RA-2 lead type elevations. ATM 200 m lead elevations are determined from ATM lead
detection analysis. A scatter plot of the ATM and RA-2 lead type elevations is shown to the
right of the floe histogram.
Collocated ASIRAS radar and ALS data in Fram Strait from ESA’s CryoSat Validation
Experiment (CryoVEx) have been used to estimate RA-2 penetration depth. However, in the
RRDP no penetration whatsoever is seen and the freeboard heights from ASIRAS and ALS
match to within few centimetres (RMSE = 2 cm), see Figure 3.5.
R = 0.99
RMSE = 0.02 m
# points = 11
y = 1.10x -0.04
R = 0.99
RMSE = 0.02 m
# points = 21
y = 1.08x - 0.05
a)
b)
Figure 3.5: Scatter plots of ASIRAS versus ALS freeboard heights from Fram Strait a) 2008
and b) 2011.
The absence of penetration depth of the ASIRAS data in the RRDP could be caused by a moist
or even wet snow surface. This is likely as the data are obtained by the end of April/beginning
of May.
28
Figure 3.6: Distribution of the 2m air temperature from JRA25 Re-Analysis data for April 24,
2008 (a) and May 3, 2011 (b), 12 UTC. The inset shows the time series of the mean
temperature of the two black boxes for the past 30 days before the experiment took place,
which is indicated by the red arrows.
In order to check the meteorological conditions during the CryoVEX expeditions with ALS and
ASIRAS usage in years 2008 and 2011 (2008: April 24, 2011: May 03) we took data of the
2m air temperature from the Japanese Re-Analysis project (JRA25) and interpolated it onto
the NSIDC polar-stereographic grid with 25 km grid resolution.
Maps of the JRA25 2m air temperature distributions for the two above-mentioned dates, 12
UTC, together with the time-series of the spatial mean 6-hourly JRA25 2m air temperature
are shown on Figure 3.4. Comparison of the situation in images a) and b) reveals that during
April 24, 2008, 2m air temperatures were well below freezing during the experiment and
during the month before. In contrast, during May 03, 2011, 2m air temperatures were higher
and still well below freezing but in 2011 the experiment took place after a period of 2m air
temperatures varying mainly between 0°C and -5°C which lasted about 2 weeks and during a
few occasions the mean 2m air temperature even exceeded the freezing point. We can
assume therefore that snow surface melting has had commenced in 2011 before the
experiment took place, modifying the snow physical properties whereas the 2m air
temperature record of 2008 suggests rather dry snow conditions devoid of surface melt.
Alternatively the absent penetration could be due to errors in the re-tracking procedure. The
method used here is based on an 80% re-tracker [Stenseng, 2012] which was chosen as it is
expected to pick up any penetration. However, the procedure tracks the first peak of the
radar signal. In case of the presence of more peaks, the first peak could in principle originate
from the snow surface whereas a second peak could originate from reflection at the snow-ice
interface. A more detailed study of the ASIRAS re-tracking procedure is beyond this study
and it is concluded that the ASIRAS combined with ALS cannot be used for this purpose, and
will not be used to draw any conclusions for the coincident RA-2 analysis.
29
R = -0.66
RMSE = 0.29 m
# points = 11
y = -0.26x + 0.22
R = 0.05
RMSE = 0.39 m
# points = 21
y = 0.02x + 0.27
Figure 3.7.: Scatter plots of RA-2 sea ice freeboard vs. ALS total (sea ice plus
snow) freeboard from Fram Strait 2008 (left) and 2011 (right)
Despite the above analysis of ASIRAS penetration, we find penetration of RA-2 when
compared to ALS snow elevations, see Figure 3.7/a. Unfortunately based on Figure 3.7/ a it is
not possible to assume that the RA-2 reflects from the snow-ice interface because: i) the
impact of the uncertainty in the retrieved snow freeboard from ALS is not considered[ ii) the
impact of the roughness is not considered; iii) the surface ice type (floe or lead) is not
considered; iv) the bias is not confirmed or compared with snow depth in the area. The
uncertainties for hfs reported, for 2009 and 2010 are up to 25cm (Figure 3.8). On Figure 3.8.
are plotted the uncertainties (from RRDP data base) of the retrieved freeboard (hf) from OIB.
a)
b)
Figure 3.8. Uncertainties of the snow freeboard from OIB; a) April, 2009; b) March-April,
2010.
If the uncertainties of hfs (hfs ) are considered no penetration or small penetration depth of
RA echo may be observed.
This requires first the ALS freeboard to be corrected with the uncertainties of the derived
freeboard from airborne laser altimeter, before any comparison of freeboard from RA and
ALS. Hendricks et al (2010) demonstrated that airborne laser and space borne radar can be
statistically biased due to presence of small patches of open water and the observed bias on
30
Figure 3.7/a could be due to impact of surface roughness or due to uncertainties in the
measured freeboard from laser altimeter but not due to penetration of RA echo in the snow.
Simulations conducted by [R30] suggest that a proportion (~7%) of the altimeter radar pulse
is reflected from the air-snow interface even for dry snow. This “early return” distorts the
returning pulse and reduces the half power time which results in an “effective scattering
surface” somewhat above the snow-ice boundary. The extent to which this occurs, assuming
that the reflectivity of the snow and ice are constant, is dependent on several parameters,
the most important of which are the snow and ice surface roughness. These quantities are
important due to their influence on the amount of scattered radiation back to the receiver.
Obtaining the snow surface roughness is relatively easy, using, for example, laser scanning
techniques on ground- or air-borne platforms but techniques to do so from satellite laser
sounding need to be established [R63]. The issue of the ice surface roughness and the
location of the dominant scattering surface are not resolved and based on laboratory
observations (Beaven et al (1995) still it is assumed that the dominant scattering surface for
RA is on the snow-ice interface.
If the radar signal does not penetrate down to the snow-ice surface, this will result in an
overestimation of RA-2 sea ice freeboard heights and consequently of sea ice thicknesses.
The difference in the shape of the echo from the ice and the leads
The process of retrieving τ from an echo involves fitting a model, which describes the echo
shape, to the data. The shape of radar altimeter echoes varies depending on the surface.
Over the consolidated ice pack, and open-ocean, diffuse echoes are observed, however over
leads the radar echoes are specular (Figure 3.9). Therefore different models are used to
retrieve τ over the ice and over the leads, which results in a bias between the elevations from
the ice floes and from the leads.
Figure 3.9. Typical ERS altimeter waveforms acquired from (a) a sea ice–covered sea
(specular waveform) and (b) open ocean (diffuse waveform).
These waveforms show received power at the altimeter versus time. Note that the y axes of
these waveforms are not to scale and that the peak power of specular waveforms can be up
to 3 orders of magnitude greater than for diffuse waveforms.
The propagation error (
) and tidal error (
)
31
The importance of these errors depends on how the difference between the ice elevation and
the ocean elevation is performed. When the ice freeboard is calculated as a difference from a
mean sea surface and mean ocean observations, then
and
is estimated by:
(3.6)
However, if there are enough elevation estimates from leads to extrapolate the ocean surface
along the satellite track, forming an ‘instantaneous’ mean sea surface then
(3.7)
The propagation, tidal and dynamic topography errors have length scales larger than typical
ice floes and therefore they cancel. The instrument errors (
) contain short scale geoid
errors, which may lead to increased estimate of the sea surface height.
The speckle error
All radar echoes exhibit a form of signal distortion know as ‘speckle’. As the speckle
decorrelates between consecutive echoes summing over n echoes reduces the noise due to
speckle by
n.
Therefore, for gridded ice thickness products, the errors depend on the
number of observations in a particular grid cell. This quantity will vary spatially due to the
convergence of the ground track at the latitudinal limit and also seasonally as the fraction of
leads and ice floes varies.
The satellite position error (
)
As with the propagation and tidal errors we expect that the orbit errors will cancel in the
freeboard calculation. Therefore
.
Sampling error (error of omission)
The radar may not sample the smallest floes and if the statistics of the sampled ice are
different to the total ice cover then this will result in an error in the spatially averaged ice
thickness. [Wingham et al., 2001] suggest that this error could be investigated using airborne
thickness measurements (laser altimetry or EM techniques) combined with imagery and by
combining satellite retrievals of sea ice thickness with imagery.
Please see [Wingham et al., 2001] for a description of the co-variance of the error in satellite
derived sea ice thickness estimates.
The re-tracking algorithms and the collocated geophysical corrections for RA-2 altimeter and
MWR data products are described in ENVISAT RA2 /MWR products manual [R24].
3.3 Errors associated with the conversion of the freeboard derived from RA to SIT
The main uncertainties of the retrieved SIT are related to uncertainties in the input variables
(e.g. snow depth and densities).
Under the assumption of hydrostatic equilibrium sea ice freeboard (hfi) can be convert to sea
ice thickness (hi) using equation (3.1).
32
In the following sub-sections we address the uncertainties: i) in the snow depth; ii) water
density; iii) ice density and the snow density.
The water density uncertainty
The water density is very close to 1024 kg/m3 and usually has a small variability compared to
the other two densities. [Wadhams, 1992] find that the density of sea water across the
Beaufort Shelf and slope off to Alaska varies between 1023.2 kg m-3 in October to 1024.2 kg
m-3 in April and a mean value of w=1024 kg m-3 is used during these months. Although
sea water density only varies in a few parts per thousand [Wingham et al., 2001], this size of
variation could result in a thickness error of the order of cm.
Ice density uncertainty
The bulk density of sea ice depends on the density of the pure ice and fractional volume of air
pockets and the amount and density of brine in the ice [Timco and Frederking, 1996]. The
brine volume generally increases with increasing temperature and when the ice is close to its
melting point the brine pockets can be large and interconnected. Therefore, sea ice is not a
close system and its density can be changed by brine drainage. It has been observed that the
density of multi-year (MY) ice is less than that of first-year (FY) ice above the water line
[Alekandrov et al., 2010], due to the fact that MY ice has undergone melt processes resulting
in air-filled pores. Figure 3.10 shows the density profile of sea ice floes in the Eurasian sector
of the Arctic Ocean [Eicken et al., 1995]. Figure 3.10 confirms that ice density is freeboard
dependent and depends on ice thickness and type. The surface layer is characterized by low
densities, corresponding to large gas bubble densities. The dashed line is a least squares fit
to the data. The profile of ice density, with mean value 887+20kg/m3 with uncertainty
40kg/m3 was measured in August –September over the Arctic Ocean [Eicken et al, 1995].
Figure 3.10. from [Eicken et al.,
1995]. Mean density profile and
standard deviation of sea ice density.
[Alekandrov et al., 2010] review variations in ice density from the literature and Server
expeditions (Figure 3.11) and estimate the sea ice density range from 720 to 940 kg/m3 with
mean density of MY ice to be 882 ± 23 kg m-3 and the bulk density of FY ice to be 916.7 ±
35.7 kg m-3. The measurements were conducted using mostly the mass/volume technique.
33
Figure 3.11: [Alekandrov et al., 2010].
A composite of sea ice density data
measurement ranges obtained from
Timco and Frederking, 1996],
The average ice density below the water line is 900 to 940 kg/m3 for both first- and multiyear ice. The sensitivity analyses and validation with ULS data demonstrated that the impact
of ice density on uncertainty of the retrieved SIT depends on algorithm for ice density used to
retrieve SIT and it is minimum when freeboard depended ice density is applied (see PVASR,
2013 report).
The presence of multiple ice types within a single footprint can result in significant error in
the estimation of the average SIT over the footprint area due to the impact of spatial and
depth variability of sea ice density on derived ice thickness. For example, [R30] simulate the
effect of the presence of a small proportion of young ice within the radar foot print and found
that the presence of highly reflective young ice results in a significant underestimate of the
average thickness.
The snow density from Warren Climatology varies seasonally but shows little geographic
variation across the Arctic [Figure 3.12]. As the seasons change from autumn to winter the
snow density increases from ~250 kg m-3 in September to ~ 320 kgm-3 in May due to the
effects of the snow settling and being packed down by the wind [Warren et al., 1999].
[Alekandrov et al., 2010] examine snow density from the Sever expeditions and find an
average snow density on first-year ice of 324 ± 50 kg m-3, which is similar to that of multiyear ice 310-320 kg m-3 [[Romanov, 1995], W99].
Table 3.1 shows the coefficients of the fit to the snow water equivalent (SWE), which is
defined as the snow depth multiplied by the snow density, the rms error in the fit and
interannual variability [Warren et al., 1999]. The distribution of SWE as a function of month
of the year, latitude (x) and longitude (y) is calculated by:
SWE = H0 + Ax + By + CxyDx2 + Ey2,
(3.8)
34
Figure 3.12. Snow density variations [R26. R19].
The coefficients of the SWE, RMS error and IAV are listed in Table 3.1 and are different
from the coefficients, used to calculate snow depth.
Table 3.1. Coefficients for the two-dimensional quadratic fit to SWE [Warren et al., 1999].
Re-distribution and compaction of snow by wind, and morphological snow cover changes
caused by moisture, heat and radiation fluxes through the snow cover usually cause a slow
increase of snow density during the freezing season. Once melt-refreeze and/or melt
commences and the snow becomes moist or even wet its density increases further.
Snow depth
The snow depth is estimated using the monthly snow depth climatology from [Warren et al.,
1999], which is based on measurements performed between 1954 and 1991 over multiyear
ice. The use of a climatology means that inter-annual and local spatial variability are not
represented. W99 provide estimates of the RMS variability, their fit to the monthly snow
depths and also their inter-annual variability (Table 3.2).
Table 3.2. Coefficients for the two-dimensional quadratic fit to snow depth from [Warren et
al., 1999].
35
The monthly snow depth hs (cm), given by two-dimensional quadratic fit to measured snow
depth is calculated by:
hs = H0 + Ax + By + CxyDx2 + Ey2,
(3.9)
where H0 is the mean monthly snow depth at the North Pole, x (latitude) and y (longitude)
are positive axes respectively along 0° and 90°E in degrees and the coefficients A,B,C,D,E
provide information for snow depth distribution and are listed in Table 3.2. The RMS error (ε)
of the fit (in cm), the slope F of the trend lines in cm yr-1, inter-annual variability (IAV) and
their uncertainty (σf) are given in Table 3.2. [Warren et al., 1999].
Higher spatial resolution snow depth estimates with low or unknown accuracy and many
restrictions for application have been available recently from AMSR-E and OIB radar.
Unfortunately the empirical AMSR-E algorithm for snow depth retrieval is with number of
limitations [Cavalieri et al, 2012] and validation with surface observations demonstrates
about 2.3 times underestimation of the snow depth, retrieved from AMSR-E microwave
observations [Worby, et al, 2008], which makes these observations not applicable for any
quantitative analyses or use for SIT retrieval from RA. The extraction of snow depth from
passive microwave observations (e.g. AMSR-E) is complicated in the Arctic by the presence of
multiyear ice, which has a radiometric signature similar to that of snow over first-year ice
[R02]. Therefore, in the Arctic, the algorithm only retrieves snow depth with great
uncertainties in the seasonal sea ice zones, i.e. where the concentration of multiyear ice is
below 20%. The snow depth from Operation IceBridge observations in the Arctic is in limited
locations and the algorithms are under development, which make them not applicable for SIT
retrieval from RA [R39, R40]. ECMWF forecasting has been also applied to calculate snow
accumulation in limited regions, starting from a certain snow depth distribution at the time of
minimum sea ice cover (e.g. mid-September), but it has not found application for SIT
retrieval from RA. Only Warren Climatology has found wide application for SIT retrieval from
ERS1,2, EnviSat and CrioSat-2 , providing snow depth and density distribution in the Arctic [
Connor et al, 2009, Laxon, et al, 2013].
3. 4. Uncertainty of the retrieved ice thickness from RA
Assuming that the uncertainties are uncorrelated, the uncertainty in sea ice thickness (to the
first order) is given by:
(3.10)
36
where
freeboard,
density,
is the uncertainty in sea ice thickness,
is the uncertainty in the sea ice
is the uncertainty in the snow depth,
is the uncertainty in the snow
is the uncertainty in the water density and
is the uncertainty in the sea ice
density [K A Giles et al., 2007]. The uncertainty in the sea ice freeboard (
) depends the
derivation of freeboard, described in section 2, as the errors in some of the parameters that
are used to calculate freeboard may cancel. For example, if one assumes that
,
,
, the error in the ice freeboard
is only dependent on the error in the first arrival time (
), which in turn depends on the
geometric, penetration and speckle errors, and the error in the short scale fluctuations in the
geoid ( ):
(3.11)
Assuming that the uncertainties of the components are not correlated and not taking into
account the error of omission the uncertainty of the gridded sea ice thickness value is
calculated considering [Bevington and Robinson, 1992] for different algorithms (input
variables of sea ice density and snow depth) and listed in PVASR report.
Because SIT, retrieved from laser and radar altimeter depends on the same input variables
(snow depth, snow, water and sea ice density) any comparison of SIT derived from RA with
SIT retrieved from laser altimeter, using different input variables may lead to misleading
results and wrong estimate of the uncertainties [Laxon et al, 2012].
37
4. Sea ice thickness, derived from alternative techniques and corresponding uncertainties.
The models and uncertainty to derive SIT from alternative technique will be discussed as part
of validation and uncertainties analyses of the retrieved SIT from RA. Upward looking sonar
(ULS) on submarine (Wadhams, et al, 1992, Rothrock et al, 2007], airborne laser altimeter [
Connor et al, 2009], SAR [Kwok et al, 1995], for thin ice, microwave radiometry (MIRAS)
aboard the Soil Moisture and Ocean Salinity (SMOS) satellite for thickness up to 50 cm
[Kaleshke et al, 2012], microwave radiometry from SSM/I or AMSR-E [Seelye et al, 2004] or
energy balance models coupled with satellite thermal infrared imagery [Wang et al, 2010].
4.1. Laser altimetry.
In contrast to radar altimetry, it is assumed that for laser altimetry (such as ICESat) the main
signal is reflected back from the air/snow interface. In common with radar altimetry, to
convert the laser measurements of snow plus ice freeboard to sea ice thickness precise
knowledge of snow, ice and water density and snow thickness is required. Therefore, the
same error sources described in §3 also apply to sea ice thickness derived from laser
altimetry measurements [ Kwok and Cunningham, 2008]. See [Kwok and Cunningham,
2008] for an example of how sea ice thickness is calculate from laser altimetry. In addition,
cloud cover affects the lasers ability to ‘see’ the surface, unlike the radar, which can ‘see’
through the clouds.
4.2. Error sources of SIT retrieved from satellite IR temperature measurements
Using a thermodynamic model [R45, R46], the SIT can be derived from satellite IR
temperatures such measured by the Pathfinder AVHRR sensor [16]. The atmospheric heat
flux algorithm used in combination with AVHRR data works for thin ice only (with thickness
less than 0.5 m, see also [R45]) and the fluxes are functions of surface skin and air
temperatures, air pressure, relative humidity, ice temperature, wind speed, cloud amount,
and snow depth. The uncertainties from the all input variables and the parameterisation of
the energy balance model affect the retrieved SIT and uncertainties can be expected to
increase with increasing SIT.
4.3. Salinity dependence of sea ice thickness
The sea ice is composed of pure ice, brine, solid salts and gas. Field measurements in the
Beaufort Sea indicate that thin ice with a thickness of a few centimetres can have a salinity
as high as 16 parts per thousand. For ice less than 0.4 m thickness, Cox and Weeks [R64]
have found an empirical linear relationship for ice salinity S (in part per thousand - ppt) and
thickness (in meter):
hit = (S – 14.24)/ 19.39 for hit<0.4m
hit = (S – 7.88)/ 1.59
(4.1)
for hit>0.4m
where the ice salinity can be calculated as a function of ice temperature derived from satellite
TIR observations (e.g. AVHRR), applying the equation derived by [R64]:
S = - 3.9921 – 22.7Ti – 1.0015Ti2– 0.019956Ti3
(4.2)
38
However, sea ice salinity does have an impact on SIT retrieval based on active and passive
microwave satellite observations via its impact on the dielectric properties of the sea ice and
snow that are relevant for its remote sensing in this frequency range. Actually, the
relationship between the sea ice salinity change with depth and its dielectric properties can
be used to retrieve thin ice thickness only for a limited thickness range (e.g. [R53], [R54],
[R55]). While the relationship can be used to retrieve the SIT, the variability of the salinity at
the same time limits the accuracy of the results.
39
5. Conclusions
This document summarise the uncertainties and methods to derive SIT from RA, and
alternative techniques. The main methods for error estimation, validate and calibrate of the
derived SIT from satellite data are grouped in : i) experimental; ii) theoretical; iii) statistical
and sensitivity analyses. Analyses of sensitivity of freeboard retrieval and freeboard-tothickness conversion algorithm to surface variables (snow depth and density, sea ice density,
sea ice type) give an estimate of the impact of input variables uncertainties on the accuracy
of the retrieved SIT.
Experimental methods involve comparison of the derived freeboard and SIT retrieved from,
e.g., radar altimetry with independent collocated freeboard and SIT measurements (satellite,
airborne, surface, underwater) at the footprint scale. Statistical, correlation and regression
analyses and comparison of the derived (gridded) sea ice product with independent SIT
products from different instruments and model simulations on the same spatial and temporal
scale as the SIT product have been widely applied for error estimation [R04-R06]. The same
spatial and temporal resolution of the data sets is required for error analyses when collocated
data sets are used. For example, within an hour, leads may open or close, deformation
features may evolve, snow might drift away and sea ice might have drifted in different
distances at the end points of a survey line, which requires precise temporal (within one
hour) and spatial collocation. A time shift of one hour between satellite data acquisition,
such as with an altimeter, and acquisition of validation data, like in-situ drilling in
combination with an over-flight of an airborne laser scanner, can be enough for adding
additional temporal and scale uncertainties. The spatial resolution and temporal sampling
of a radar altimeter, the minimum number of altimeter measurement samples required to
reduce the noise to a reasonable level, the high spatial variability of the SIT and also the
minimum number of consecutive measurements to obtain a representative SIT estimate,
have to be considered when collocated data are used for uncertainty analyses.
40