ArticlePDF Available

Arctic Ocean Sea Level Record from the Complete Radar Altimetry Era: 1991–2018


Abstract and Figures

In recent years, there has been a large focus on the Arctic due to the rapid changes of the region. Arctic sea level determination is challenging due to the seasonal to permanent sea-ice cover, lack of regional coverage of satellites, satellite instruments ability to measure ice, insufficient geophysical models, residual orbit errors, challenging retracking of satellite altimeter data. We present the European Space Agency (ESA) Climate Change Initiative (CCI) Technical University of Denmark (DTU)/Technischen Universität München (TUM) sea level anomaly (SLA) record based on radar satellite altimetry data in the Arctic Ocean from the European Remote Sensing satellite number 1 (ERS-1) (1991) to CryoSat-2 (2018). We use updated geophysical corrections and a combination of altimeter data: Reprocessing of Altimeter Product for ERS (REAPER) (ERS-1), ALES+ retracker (ERS-2, Envisat), combination of Radar Altimetry Database System (RADS) and DTUs in-house retracker LARS (CryoSat-2). Furthermore, this study focuses on the transition between conventional and Synthetic Aperture Radar (SAR) altimeter data to make a smooth time series regarding the measurement method. We find a sea level rise of 1.54 mm/year from September 1991 to September 2018 with a 95% confidence interval from 1.16 to 1.81 mm/year. ERS-1 data is troublesome and when ignoring this satellite the SLA trend becomes 2.22 mm/year with a 95% confidence interval within 1.67–2.54 mm/year. Evaluating the SLA trends in 5 year intervals show a clear steepening of the SLA trend around 2004. The sea level anomaly record is validated against tide gauges and show good results. Additionally, the time series is split and evaluated in space and time.
Content may be subject to copyright.
remote sensing
Arctic Ocean Sea Level Record from the Complete
Radar Altimetry Era: 1991–2018
Stine Kildegaard Rose 1,*,† , Ole Baltazar Andersen 1, Marcello Passaro 2,
Carsten Ankjær Ludwigsen 1and Christian Schwatke 2
1Technical University of Denmark—National Space Institute (DTU Space), 2800 Kgs. Lyngby, Denmark
2Deutsches Geodätisches Forschungsinstitut der Technischen Universität München (DGFI-TUM),
80333 Munich, Germany
*Correspondence:; Tel.: +45-45259742
Current adress: Department of Geodesy, DTU Space—National Space Institute, Elektrovej Build. 228,
2800 Kgs. Lyngby, Denmark.
Received: 12 June 2019; Accepted: 11 July 2019; Published: 14 July 2019
In recent years, there has been a large focus on the Arctic due to the rapid changes
of the region. Arctic sea level determination is challenging due to the seasonal to permanent
sea-ice cover, lack of regional coverage of satellites, satellite instruments ability to measure ice,
insufficient geophysical models, residual orbit errors, challenging retracking of satellite altimeter
data. We present the European Space Agency (ESA) Climate Change Initiative (CCI) Technical
University of Denmark (DTU)/Technischen Universität München (TUM) sea level anomaly (SLA)
record based on radar satellite altimetry data in the Arctic Ocean from the European Remote Sensing
satellite number 1 (ERS-1) (1991) to CryoSat-2 (2018). We use updated geophysical corrections and
a combination of altimeter data: Reprocessing of Altimeter Product for ERS (REAPER) (ERS-1), ALES+
retracker (ERS-2, Envisat), combination of Radar Altimetry Database System (RADS) and DTUs
in-house retracker LARS (CryoSat-2). Furthermore, this study focuses on the transition between
conventional and Synthetic Aperture Radar (SAR) altimeter data to make a smooth time series
regarding the measurement method. We find a sea level rise of 1.54 mm/year from September 1991 to
September 2018 with a 95% confidence interval from 1.16 to 1.81 mm/year. ERS-1 data is troublesome
and when ignoring this satellite the SLA trend becomes 2.22 mm/year with a 95% confidence interval
within 1.67–2.54 mm/year. Evaluating the SLA trends in 5 year intervals show a clear steepening of
the SLA trend around 2004. The sea level anomaly record is validated against tide gauges and show
good results. Additionally, the time series is split and evaluated in space and time.
radar altimetry; satellite altimetry; arctic ocean; remote sensing of the oceans; sea level
rise; polar area
1. Introduction
The Arctic region has warmed faster than any other parts of the Earth, where the sea level of the
Arctic Ocean is an important climate indicator. The arctic sea-ice is decreasing, and has since 1997
experienced a steepening in the decrease [
]. In the fifth Intergovernmental Panel on Climate Change
(IPCC) report a global total sea level rise of 2.8
0.7 mm/year in the period of 1993–2010 is found [
The sea level rise is due to: (1) Thermal expansion [
]. 93% of the atmospheric energy imbalance,
which is caused by greenhouse gases, accumulates in the ocean as ocean heat content. Recent models
show an increasing ocean warming trend in the upper 2 km of the oceans [
]. (2) Land water storage
from human interactions i.e., ground water depletion and reservoir storage [
]. (3) Glacier and ice
Remote Sens. 2019,11, 1672; doi:10.3390/rs11141672
Remote Sens. 2019,11, 1672 2 of 29
sheet mass losses. Outlet glaciers are losing mass more rapidly [
], contributing to the sea level rise,
changing the oceans freshwater flux, and influencing the ocean thermohaline circulation [8].
The polar oceans are often not included in the global sea level estimations and can be seen
as white spots on the global sea level maps. This is because of the challenging polar sea level
determination due to; the seasonal to permanent sea-ice cover, the lack of regional coverage of
satellites, satellite instruments ability to measure ice, insufficient geophysical models, residual orbit
errors and retracking of satellite altimeter data.
The sea-ice cover is in constant change. The sea-ice extent is the largest in March and the smallest
in September. The Norwegian and Barents Sea are only seasonally covered by sea-ice while the
central part up to the Canadian Archipelago and the North coast of Greenland are permanently ice
covered (see Figure 1for an Arctic Ocean overview). The older ice is pushed against these parts,
and additionally, the Canadian Archipelago and the land-fast ice areas are also the part with the fewest
leads and consequently the most inaccurate sea level determination [9].
Figure 1.
Overview map of the Arctic Ocean. The map show the tide gauges (red dots) used to
validate the SLA and the different sectors (divided in blue punctured lines) used to investigate the
different Arctic regions. The four regions are: I: Fram Strait, Greenland Sea, Norwegian Sea and
Barents Sea. II: The Russian Arctic: Kara Sea and Laptev Sea, III: East Siberian Sea and Beaufort Sea,
IV: The Canadian Archipelagos and Baffin Bay.
Sea-ice affects the returned satellite radar signal (or waveform) resulting in a poorer coverage
and a lower quality of the return signal. Sea level estimates in the sea-ice covered areas are dependent
on gaps in between ice floes (leads or polynyas). From now on we are not separating between leads
and polynyas but referring to leads as ocean water surrounded by frozen ice. Leads are often very
flat ocean surfaces, where there are almost no scatter from the radar wave. This will be registered as
a very peaky waveform in the received echo. If leads are located off-nadir their strong backscatter can
substantially decrease the quality of the range retracking, this is also known as snagging [
]. In case of
Envisat, the nominal circular footprint of 2 km in diameter [
] can increase up to 10 km [
] for strong
off-nadir backscatter sources. Despite its much smaller along-track footprint (1.65 km
0.30 km),
CryoSat-2 can also be affected by off-nadir leads, which will result in erroneous range estimates [
Refrozen leads are often seen as normal specular lead waveforms but they can be biased up to a couple
of centimeters. Another source of errors can be melt ponds on-top of the ice in Spring/Summer and
ice freeze-ups in Autumn.
Remote Sens. 2019,11, 1672 3 of 29
Also the geophysical range corrections are less accurate in the Arctic due to the sea-ice
contamination of the radiometer and the lack of observations. The tides and the inverse barometer
effect (IBE) are the most important parameters, but also the most uncertain [9,14].
The satellite altimeter era has completed more than 25 years of measurements distributed over
several satellites: ERS-1, ERS-2, Envisat, CryoSat-2, ICESat-1/2, SARAL and Sentinel-3A/B. All these
satellites have a geographical coverage suitable for Arctic Ocean research. The T/P and Jason satellite
series have proven worthy for mid latitude SLA studies, but do not cover the Arctic region. In this
study, we use data from four ESA radar altimeter satellites; ERS-1, ERS-2, Envisat and CryoSat-2. In the
earlier altimeter satellite missions (ERS-1, ERS-2 and Envisat) orbit errors up to 5 cm still exists [15].
The Arctic Ocean is lacking in-situ measurements consistent in time and space, mainly due to its
harsh environment. Various publications (e.g., [
]) of the Arctic sea level from tide gauge data
exists, where the sea level are measured along the Russian and Norwegian coasts. There exist few tide
gauges in the interior of the Arctic Ocean, and they are all short time series. Several bouys have been
deployed in the Arctic Ocean ex. Argo (, Ice-Tethered Profiler (
/website/itp), UNCLOS and GreenArc [
] The Argo buoys have shown great results in validating
altimeter data [19], but are not yet densely deployed in the Arctic Ocean.
The first Sea Surface Height (SSH) studies covering large parts of the Arctic Ocean were computed
from the ERS satellites to produce sea-ice thicknesses [
] and gravity anomalies [
]. Peacock and
were the first to construct an ocean product, a mean SSH (MSS), from the Arctic Ocean
using ERS-1 and ERS-2 altimeter data from a 10 year period. Since then several e.g., [
] have
followed. Global MSS products are available from: CNES/CLS (not covering the Arctic) [
], DTU [
In this paper, as part of ESA’s Sea level CCI (SL_CCI) and the Sea Level Budget Closure
(SLBC_CCI), we use 27 years of radar satellite altimeter data for constructing a new improved monthly
sea level record for the Arctic Ocean - the CCI DTU/TUM Arctic Ocean data set. We find a sea level
rise of 1.54 mm/year of the Arctic Ocean covering 65
N to 81.5
N latitude and
to 180
from September 1991 to September 2018, with a 95% confidence interval of 1.16–1.81 mm/year.
The coverage from the ERS-1 satellite is sparse during periods of time and the time series may be more
error prone in this period, therefore looking at the time series starting from ERS-2 we get a sea level
rise of 2.22 mm/year with a 95% confidence interval within 1.67–2.54 mm/year.
The paper starts by describing the data used (Section 2). We use a combination of tailored
level-2 (L2) ERS-1 data together with a new retracking of ERS-2 and Envisat data and with
a combination of state-of-the-art altimeter data and retracked data from CryoSat-2. In Section 3,
the methods are described in making the SLA product from pre-processing (Section 3.1) and
geophysical corrections (Section 3.2) to handling the sea-ice (Section 3.3) and the intermission
biases (Section 3.4) between the different satellites. The SLA product resampling and gridding are
described (Section 3.6), and finally, in Section 3.7 a bootstrap analysis is described to evaluate the SLA
uncertainties. In Section 4, the results are described and validated. The section starts by showing
the resulting SLA uncertainty (Section 4.1). The results are described as regional trends (Section 4.2),
inter-annual variability (Section 4.3) and regional variability (Section 4.4). The sea level anomalies
(SLA) are validated against six tide gauge stations shown in Figure 1. This is described in Section 4.5.
The results are discussed in Section 5and summarized in the conclusion (Section 6).
2. Data
This section describes the data used in this study.
2.1. Altimetry Data
The CCI DTU/TUM Arctic SLA contains data from four ESA radar altimeter satellites ERS-1,
ERS-2, Envisat, CryoSat-2. ERS-1, ERS-2 and Envisat are conventional altimetry or low resolution
mode (LRM) data sets processed with a single processor, while CryoSat-2 consists of three types: LRM,
Remote Sens. 2019,11, 1672 4 of 29
Synthetic Aperture Radar (SAR) and SAR Interferometry (SARIn), which are processed with different
processors. For satellite specific details see Appendix A. In Section 3.2 the geophysical range correction
data are described.
2.2. Ice Concentration Data
In Section 3.3, we use sea-ice concentration data in separating sea-ice data from ocean data.
The sea-ice concentration data are derived in an operational product (after 2015) [
] and a reprocessed
product (before 2015) [
] by the EUMETSAT Ocean and Sea Ice Satellite Application Facility.
Both products are given as sea-ice concentrations in 10 km Polar Stereographic grids for every six hours.
2.3. Tide Gauge Data
In validation of the CCI DTU/TUM SLA data set (Section 4.5), tide gauge data from the Permanent
Service for Mean Sea Level (PSMSL) [
] are used. The tide gauge data are given as monthly SLAs.
Six tide gauges are chosen spread along the coast of the Arctic Ocean (Figure 1).
3. Generation of the Sea Level Product
The Arctic Ocean SLAs are computed by the following steps:
1. Pre-processing
2. Adding/removing geophysical corrections
Sea-ice concentration data are used to discriminate between the sea-ice cover and the open ocean
4. Threshold criterias are used to separate the leads/open ocean from the sea-ice
5. Inter-satellite biases are determined and corrected
6. Removing outliers
7. Resampling and gridding the data to compute the final Arctic SLA
8. Uncertainty analysis
3.1. Pre-Processing
Pre-processing details for the individual satellites are described in Appendix A.
3.2. Geophysical Corrections
The geophysical corrections were updated to get a more uniform product, suitable to compare the
SLAs in between satellites. Table 1summarizes the corrections used.
Table 1.
Data origin and applied geophysical corrections. O, L, LP tides are the Ocean tide, ocean
loading tide, long-periodic non-equilibrium ocean tide, LP otide + setide includes the long-periodic
ocean tide and the solid earth tide.
ERS-1 ERS-2 Envisat CryoSat-2 [33]
Data origin REAPER L2 [34] ALES+ [34,35] ALES+ [35,36] LARS/RADS [33,37,38]
Wet troposphere ECMWF [39] ECMWF [39] ECMWF [39] ECMWF [39]
Dry troposphere Radiometer/ECMWF [39] Radiometer/ECMWF [39] ECMWF [39] ECMWF [39]
Ionosphere NICO [40]/GIM [41] NICO [40]/GIM [41] Doris [36] GIM [41]/Bent [42]
DAC ERA-Interim [43] ERA-Interim [43] ERA-Interim [43] DAC-ECMWF [44]
O, L, LP tides FES2014 [45] FES2014 [45] FES2014 [45] FES2014 [45]
LP otide + setide Cartwright [46] Cartwright [46] Cartwright [46] Cartwright [46]
Pole tide Wahr [47] Wahr [47] Wahr [47] Wahr [47]
Sea state bias Altimetrics [34] ALES+ [48] ALES+ [48] None/RADS [38]
Mean sea surface DTU18 [49] DTU18 [49] DTU18 [49] DTU18 [49]
The preferred method for estimating the wet tropospheric correction over the Arctic Ocean is to
use modeled data, due to the radiometer contamination by the sea-ice [
]. For most of the satellites
a model correction is available. For ERS-1 REAPER data the microwave radiometer wet tropospheric
Remote Sens. 2019,11, 1672 5 of 29
correction is applied over the ocean if valid or else a model correction is applied. The authors were
not aware of a way to see which of the corrections were applied, and therefore it was not possible to
change this correction.
We use the FES2014 [
] ocean tide model with loading effects. This model is optimized in the
Arctic Ocean compared to previous versions. The tide model is limited in coastal areas resulting in
a final data set not defined close to the coast. FES2014 was produced by Noveltis, Legos and CLS Space
Oceanography Division and distributed by Aviso, with support from CNES (http://www.aviso.altime
The atmospheric correction in the Arctic is very important since amplitudes of the signal can
reach 1 m, i.e., greater than the SLA signal. Normally in the Arctic, IBE is favored over the Dynamic
Atmosphere Correction (DAC) including high atmospheric fluctuations, because of high latitude issues.
For consistency, ERS-1, ERS-2 and Enivsat are reprocessed with the DAC ERA-Interim [
] by linear
interpolation in space and time. In consequence, this will give more outliers in the data. The DAC
ERA-Interim product are computed in the period of 1991–2015, not covering the total CryoSat-2 period.
Therefore, the DAC-ECMWF [
] from CLS is here used from the CryoSat-2 GDR product. Various
models were tested, and this was proven to be the one closest to the DAC ERA-Interim model.
The applied sea state bias correction for ERS-1 is taken from the REAPER product. For ERS-2 and
Envisat the sea state bias is derived from the ALES+ retracker and applied at 20 Hz [
]. For CryoSat-2
only sea state bias for the LRM mode is applied, which is a hybrid sea state bias from the RADS
product. For most cases it is fair to ignore the sea state bias in SAR and SARIn mode, such leads are
very flat surfaces, where the sea state bias is very close to zero.
The DTU18 MSS was used as a reference [
]. The new MSS from DTU is improved in the central
Arctic region and in coastal zones. It has a bias towards recent years sea level heights including three
years of Sentinel-3A and eight years of improved CryoSat-2 data.
3.3. Lead and Ocean Discrimination
The Arctic Ocean SLA record is derived by separating leads in the sea-ice cover and open ocean
according to the different classification of their surfaces. Various sea-ice types can mistakenly be
associated with open ocean waveforms. The ocean is separated from the sea-ice cover by the ice
concentration grids (Section 2.2). For more details see Appendix B.
Sea-ice and mixed surfaces are removed by using the waveform Pulse Peakiness (PP) and the
width of the leading edge. Furthermore, for CryoSat-2, the stack standard deviation is used to identify
the leads. For removing erroneous data in the open ocean (that could be data from the ice edge or
near the coast), the PP and the backscatter coefficient are used. There exists many variations of the
PP formula e.g., [
]. The values used in this study for each satellite are shown in Table 2.
All references to the PP (in Table 2and in text) are described as in [
], which is given by the waveform
maximum power received multiplied with the sum of all range bin powers. In parenthesis, PP values
are described as in [
] which is the same formulation as [
] but multiplied with a constant of 31.5.
This formulation was first used for the ERS satellites. The choice of these threshold values are based
on several studies [11,13,22,53,54] and adjusted and evaluated for this study.
Remote Sens. 2019,11, 1672 6 of 29
Table 2.
Thresholds used in lead and ocean discrimination. The table columns are PP, stack standard
deviation (St. Std), and width for lead discrimination and PP. Backscatter coefficient (
) for the ocean
discrimination. The PP values are described as in [
] and in parenthesis as in [
]. The PP is calculated
differently for CryoSat-2 (see the details in the text). The two numbers corresponds to SAR and SARIn,
respectively. The width in ERS-1 is from the REAPER product and is the OCOG width, for ERS-2 and
Envisat the width is the ALES+ leading edge rising time, and for CryoSat-2 the width is the width of
the Gaussian fit.
Lead Ocean
PP >St. Std Width <PP <σ0<
ERS-1 0.60 (19) - 3 0.048 (1.5) 15
ERS-2 0.65 (20.5) - 3 0.048 (1.5) 15
Envisat 0.71 (22.5) - 3 0.048 (1.5) 15
CryoSat-2 0.35/0.25 (11/7.9) 4 0.9 - -
3.4. Intermission Bias
To get a seamless transition between conventional altimetry (from ERS-1/2, Envisat, CryoSat-2
(LRM)) and SAR/SARIn (CryoSat-2) altimetry can be error prone, especially in the Arctic due to
the different data coverage. SAR altimeter data have much more data over the sea-ice cover, while
conventional altimetry are having troubles. Conventional and SAR/SARIn altimetry data sets are
covering different regional areas and are processed with different strategies and having different
retracking corrections.
For CryoSat-2, the best approach of merging the different satellite measurement types (LRM, SAR
and SARIn) has proven to be a detailed study of individual satellite tracks (not shown). RADS data are
in LRM while LARS data are covering SAR and SARIn, so no data are overlapping in time. We found
a retracker bias between RADS and LARS of 12.9 cm.
The transition between the four satellite missions, the intermission biases were estimated and
minimized. The following steps were completed to handle the intermission biases:
Monthly medians were calculated for each mission, over the entire Arctic Ocean, covered by the
data sets
For overlapping mission pairs (either ERS-1 and ERS-2, ERS-2 and Envisat, or Envisat and
CryoSat-2) coinciding months (only full months considered) were detected and extracted
3. For each overlapping pair, the median difference was calculated and the data sets were aligned
The biases between the satellites are: ERS-1/ERS-2
0.67 m, ERS-2/Envisat
0.53 m and
Envisat/CryoSat-2 0.03 m
Figure 2shows the monthly median of each overlapping satellite pair. The Pearsons correlation
coefficient of the three satellite pairs ERS-1/ERS-2, ERS-2/Envisat and Envisat/CryoSat-2 gives 0.52,
0.96, 0.95, respectively.
(a) ERS-1 and ERS-2 (b) ERS-2 and Envisat
Figure 2. Cont.
Remote Sens. 2019,11, 1672 7 of 29
(c) Envisat and CryoSat-2
Figure 2.
Monthly median of the entire Arctic in the overlapping periods for (
) ERS-1 (e1) and
ERS-2 (e2), (
) ERS-2 and Envisat (n1) and (
) Envisat and CryoSat-2 (c2). In the top right corner of
each figure the correlation coefficient is shown.
3.5. Removing Outliers
The outlier removal is carried out in two steps. First, as mentioned in Appendix A, outliers
are removed from each track with a MAD outlier detector to get rid of the largest outliers. Second,
outliers are detected and removed on a monthly basis with a hard cut-off of
0.3 m from the median.
This was done similar to Cheng et al.
. The hard cut-off resulted in rejection of 18.05% data for
ERS-1, 2.45% data for ERS-2, 0.52% for Envisat and 0.06% of data for CryoSat-2. The large removal
of ERS-1 data are due to error-prone orbit estimation and bad data sampling, which are causing bad
waveforms, resulting in wrong height estimates.
3.6. The Arctic Sea Level Anomaly Product
First, monthly data are averaged in cells of 0.2
to overcome the sampling dissimilarity
in latitude, which would favor high latitude data especially for Cryosat-2, where the data coverage
is much larger than for the conventional altimetry satellites. Second, a least squares collocation
with second-order Markov covariance function [
] is used to grid the monthly data. The final grid
size is 0.25
latitude by 0.5
longitude using a 500 km correlation length with a RMS noise of 2 cm.
The outputs from the collocation are the SLA data record and a interpolation error estimate both given
in monthly grids from September 1991 to September 2018, covering 65
and 180
in gridline registration. The mean SLAs are shown in Figure 3for each satellite: ERS-1 (a), ERS-2 (b),
Envisat (c) and CryoSat-2 (d). The mean SLA is slightly higher over the sea-ice cover. This is especially
the case for ERS-1. For ERS-2 and CryoSat-2 low SLAs controls the Canadian Arctic and the Beaufort
Gyre areas, while we see a large mean SLA for Envisat in the Kara Sea. The SLAs in the Fram Strait
and Barents Sea areas are slightly negative for the three first missions, while it is slightly positive for
CryoSat-2. These figures will be discussed further in Section 5.1.
(a) (b)
Figure 3. Cont.
Remote Sens. 2019,11, 1672 8 of 29
(c) (d)
Figure 3.
The average SLA in meters for each satellite period: (
) ERS-1, (
) ERS-2, (
) Envisat,
(d) CryoSat-2.
3.7. Uncertainty Estimates
In Section 1, multiple error sources that contribute to the total uncertainty of the derived SLA are
introduced. These are errors on the altimeter instruments, the orbit determination, the retracking of
the radar signal, and from this follows the many uncertainties on the geophysical range corrections
(Section 3.2). We can now calculate the true SLA including noise. The exact size of this noise coming
from the uncertainties described above are not known, but Ablain et al.
looked into this error
budget. On top of all these uncertainties there can be errors in the discrimination of ocean and leads
(Section 3.3), inter-satellite biases (Section 3.4), in making of the SLA grids (Section 3.6), in making of
the total SLA time series and trend maps (Section 4.2), and furthermore, uncertainties can arise from:
retracker biases, interpolation, filtering, sampling. The size of all these individual uncertainties are,
however, not well known, and additionally it is difficult to propagate the uncertainties analytic in
the long processing chain. As an alternative we apply a bootstrap approach [
] to estimate the error
of the SLA. Bootstrapping embrace all the variations from the various uncertainties. To obtain valid
error estimates using bootstrap, the observations must be independent and the bootstrap data sets
must resemble the original data set. Hence, to better approximate independent observations, a block
bootstrap is used.
The specific bootstrap procedure to derive the error for each monthly data set is carried out as
follows: (1) the data are split in
non-overlapping blocks. (2) 1000 bootstrap realizations are created,
by sampling with replacement among the blocks. (3) For each bootstrap data set the SLA is derived in
the same way as described in Section 3.6. (4) Finally we have 1000 estimates of the SLA for each grid
cell from which we can extract error information such as standard deviation and confidence interval.
In Appendix Ca more thorough review of the bootstrapping procedure is described.
It is only valid to show results with a standard deviation if the results are normal distributed.
The Arctic SLA distributions are not normal distributed for all grid cells in the Arctic Ocean
(see Figure A1 in Appendix Cfor more details). Therefore, the uncertainty is expressed in a 95%
confidence level. The results are shown with the median and not the mean value, because of the
skewness of the distributions (Figure A1, Appendix C).
4. The Arctic Sea Level Anomaly Record
The resulting CCI DTU/TUM Arctic SLA product is analyzed in this section. The SLA product is
given by monthly grids from September 1991 to September 2018. These grids are available at DTU: and at ESA SLBC_CCI:
Firstly, the total uncertainty of the SLA product is shown. Secondly, we investigate the spatial
trend patterns over the entire Arctic region. Thirdly, we show the averaged inter-annual variability.
Fourthly, the SLA is validated against tide gauges.
Remote Sens. 2019,11, 1672 9 of 29
4.1. Uncertainty of the Arctic Sea Level Product
The total uncertainty of the Arctic CCI DTU/TUM sea level product from the bootstrapping is
summarized (Figure 4). This is given by the monthly median ranges of the 95% confidence interval
(i.e., the median of the SLA range between the percentiles 2.5% and 97.5%) from September 1991 to
September 2018. We see larger uncertainties in the interior of the Arctic where permanent and seasonal
sea-ice appears with a SLA range 50–60 mm compared to the ice-free regions with a SLA range of
10–20 mm.
Figure 4.
The total uncertainty of SLAs from the bootstrapping, given as the median SLA range
between the 2.5% percentile to the 97.5% percentile (i.e., the 95% confidence level interval) of monthly
data in the SLA product from September 1991 to September 2018.
4.2. Regional Trends in the ERS-2 to CryoSat-2 Era
We investigate the spatial trend pattern from 65
N to 81.5
N in the entire Arctic Ocean.
In Figure 5a the spatial pattern is shown covering the time period from January 1996 to September 2018.
Here, the ERS-1 data are dismissed due to too low data distribution in the Eastern sector. Furthermore,
all data are eliminated with an interpolation error (from the collocation) above 10 cm. We find a pattern
with a high trend >10 mm/year in the Beaufort Gyre, a slightly negative trend or no trend in the
Russian sector (
2 to 1 mm/year), trends between 3–7 mm/year in the Barents Sea and in the Fram
Strait, and a strong negative trend in the northern Baffin Bay. The regional trend uncertainties are
shown as the 2.5% percentile (Figure 5b) and the 97.5% percentile (Figure 5c) corresponding to the
95% confidence level.
Remote Sens. 2019,11, 1672 10 of 29
(a) SLA Trend
(b) SLA trend of percentile 2.5%. (c) SLA trend of percentile 97.5%.
Figure 5.
) The CCI DTU/TUM SLA trends from January 1996 to September 2018 given in mm/year.
) show the SLA trend uncertainty in the same period. There is found a 95% confidence interval of
the SLA trend within (b,c).
4.3. Inter-Annual Variability
The SLA data are averaged for each month with a cosine latitude weighting (Figure 6).
The seasonal variability are plotted from September 1991 to September 2018 (27 years) in (Figure 6a)
and January 1996 to September 2018 (almost 23 year) in (Figure 6b), respectively. We are investigating
the time series with and without ERS-1, because the coverage of the ERS-1 satellite is sparse during
periods of time (especially in the ice covered regions), and therefore the time series may be more error
prone in this period. Both figures show solutions with and without Glacial Isostatic Ajustment (GIA).
The applied GIA model is from Caron et al.
, which is kindly converted to sea level anomalies and
Remote Sens. 2019,11, 1672 11 of 29
associated standard deviations by Benjamin D. Gutknecht. Generally, we see a seasonal variability
of high sea level in late Autumn and a low sea level in the Spring. Both time series have a positive
trend with a sea level rise of 1.54 (1.40) mm/year and 2.22 (2.08) mm/year in the respectively periods
with and without (in parenthesis) GIA correction). There is a 95% confidence that data lies within
1.16 (1.01)–1.81 (1.67) mm/year and 1.67 (1.52)–2.54 (2.40) mm/year, respectively.
(a) Monthly SLA values from September 1991 to September 2018.
(b) Monthly SLA values from January 1996 to September 2018.
Figure 6.
Monthly SLA values. (
) From September 1991 to September 2018 with a linear trend of
1.54 and 1.40 mm/year with a 95% confidence level of data laying within 1.16 to 1.81 and 1.01 to
1.67 mm/year with and without GIA correction, respectively. (
) From January 1996 to September 2018
with a linear trend of 2.22 and 2.08 mm/year with a 95% confidence level of data laying within 1.67 to
2.54 mm/year and 1.52 to 2.40 mm/year with and without GIA correction, respectively. The blue and
yellow shadows are the 95% confidence level for measurements with GIA and without GIA, respectively.
The uncertainties expressed in Figure 6as light yellow (no GIA) and light blue (GIA) shadows are
derived by continuing each of the 1000 bootstrap realizations through the same procedure as described
in Section 3.6. The uncertainties are given as the median SLA range of the 1000 bootstrap realizations
in the 95% confidence level.
4.4. Regional Sea Level Variability
In Figure 7, the CCI DTU/TUM sea level record from 1996–2018 is divided into four sectors
(Figure 1). The regional SLAs with and without GIA and the associated uncertainties are summarized
in Table 3.
Remote Sens. 2019,11, 1672 12 of 29
Table 3.
SLA trend for each sector and the associated uncertainty. The SLA trend is given with and
without GIA correction and the uncertainty is given by a 95% confidence level.
SLA Trend (No GIA) 95% Conf. Level (No GIA) SLA (GIA) 95% Conf. Level (GIA)
mm/year mm/year mm/year mm/year
Sector I 3.04 2.96–3.23 3.19 3.10–3.37
Sector II 0.33 0.58–1.28 0.04 0.86–1.00
Sector III 4.06 2.41–4.71 5.77 4.12–6.42
Sector IV 0.49 0.72–1.15 0.63 1.84–0.03
Two areas (Sector I (Figure 7a) and III (Figure 7c) have a clear sea level rise in the period.
The maximum SLA trend is observed in the Beaufort Gyre (Sector III) up to approximately 10 mm/year.
In Sector I (Figure 7a) the highest SLA trend is observed in the southern part towards the Norwegian
coast, and smallest along the coast of Greenland and in the upper northeastern part.
Considering the confidence level, Sector II has no or a little positive/negative trend. Sector IV has
a positive trend when no GIA is applied, but a negative trend when it is applied. The most negative
trend is in the northeastern Baffin Bay of about
10 mm/year. It is unclear if this is due to fresh water
flow from the large outlet glaciers or a simple artifact of the LRM to SARIn transition.
1996 1998 2000 2002 2004 2006 2008 2010 2012 2014 2016 2018
Time (year)
Sea level (cm)
(a) Sector I
1996 1998 2000 2002 2004 2006 2008 2010 2012 2014 2016 2018
Time (year)
Sea level (cm)
(b) Sector II
1996 1998 2000 2002 2004 2006 2008 2010 2012 2014 2016 2018
Time (year)
Sea level (cm)
(c) Sector III
Figure 7. Cont.
Remote Sens. 2019,11, 1672 13 of 29
1996 1998 2000 2002 2004 2006 2008 2010 2012 2014 2016 2018
Time (year)
Sea level (cm)
(d) Sector IV
Figure 7.
Regional changes of sea level in the different sectors shown in Figure 1. Blue is the monthly
CCI DTU/TUM SLA in meters and the red line is the estimated trend in mm/year for the given
sector. The 95% confidence level is given by the blue and yellow shadows with GIA and without
GIA, respectively.
4.5. Validation
Tide gauges (Section 2.3) in the Arctic are sparsely distributed and gauges with long time series are
rare. The six tide gauges used in this study (Figure 1) are chosen due to their geographical distribution
in the region, their time span covering most possible of the altimetry era and their continuity in time.
The tide gauges are mounted on land and do not account for GIA effects nor the atmospheric
loading. Consequently, in the comparison, the atmospheric loading is not applied to the altimetry data.
There is a large GIA signal in the Arctic, with large variations over the region, but GIA models are very
uncertain [
]. The most direct method for determining the local vertical displacement is by GPS
measurements. The GPS vertical displacement includes both the GIA and the elastic signal, whereas
the elastic displacement comes from present displacements as mass changes from ex. outlet glaciers or
ground water depletion. The elastic displacement is very small if the tide gauge is far from the large
mass changes. A vertical GPS displacement is provided when available, or else the closest grid point
from the Caron et al. [58] GIA model (given with two times standard deviation) is used (See Table 4).
Table 4. Vertical displacement from from GPS and the Caron et al. [58] GIA model.
Tide Gauge Vert. Disp. GIA
mm/year mm/year
Ny Ålesund 7.98±0.49 10.47 ±0.67
Honningsvåg 1.9±0.3 21.344 ±0.42
Prudhoe bay - 1.51 ±0.095
Vise Ostrov - 1.96 ±0.38
Golomianyi Ostrov - 1.99 ±0.33
Sannikova Proliv - 0.48 ±0.21
1Obtained dec. 2018 from [61]; 2[62].
The results of the comparison between the CCI DTU/TUM Arctic SLA and the tide gauge data
are shown in Figure 8and in Table 5and described with more details in Appendix D. We are using an
inverse distance weighted average of data in a radius of 350 km from the tide gauge station. Using a
radius of 350 km is also done in [24].
Figure 8compares each satellite relative to the tide gauge. The tide gauge is shown with an orange
curve, while the altimetry data are shown with different colors depending on the satellite: ERS-1
(red), ERS-2 (blue), Envisat (green) and CryoSat-2 (Grey). Also trend lines for each satellite are shown
for both tide gauges and altimetry data. In Figure A2, Appendix D, the total time series is shown
together with the corresponding trend line for every tide gauge station. The trend differences between
the altimetry and tide gauge data for Sannikova Proliv (Figure A2e) and Prudhoe Bay (Figure A2f)
are large.
Remote Sens. 2019,11, 1672 14 of 29
Figure 8. Cont.
Remote Sens. 2019,11, 1672 15 of 29
Figure 8.
Tide gauge comparison of the six tide gauges for each satellite period, where red (ERS-1),
blue (ERS-2), green (Envisat) and grey (CryoSat-2) are the altimetric data and orange is the tide gauge
data. The tide gauges are evaluated relative to each individual satellite and accordingly a gap appear in
between the satellites where data overlaps. Also trend lines are shown on the figures. (
) Honningsvåg.
(b) Ny Ålesund. (c) Vise Ostrov. (d) Golomianyi Ostrov. (e) Sannikova Proliv. (f) Prudhoe bay.
Table 5.
Tide gauge comparisons. The second column shows the number of months analyzed.
The second part of the table summarizes the tide gauge comparison given as the RMSE (in meters) with
Persons’ correlation coefficient in parenthesis for ERS-1 (E1), ERS-2 (E2), Envisat (N1), CryoSat-2 (C2),
before GIA correction, in the period starting from 1996 (without ERS-1) and for the total time period.
Tide Gauges No. of Month RMSE (R)
E1 E2 N1 C2 Pre GIA 1996– Total
Ny Ålesund 312 0.080
(0.49) 0.031
(0.92) 0.034
(0.90) 0.038
(0.91) 0.072 (0.70) 0.042
(0.88) 0.050
Honningsvåg 316 0.089
(0.81) 0.037
(0.97) 0.032
(0.97) 0.060
(0.91) 0.057 (0.91) 0.047
(0.94) 0.055
Prudhoe bay 273 0.11
(0.15) 0.099
(0.57) 0.15
(0.20) 0.072
(0.91) 0.12 (0.53) 0.11
(0.57) 0.12
Vise Ostrov 221 0.12
(0.51) 0.10
(0.59) 0.097
(0.55) - 0.11 (0.52) 0.098
(0.57) 0.10
Golomianyi Ostrov 202 0.099
(0.16 ) 0.086
(0.49) 0.077
(0.66) - 0.085 (0.53) 0.081
(0.59) 0.081
Sannikova Proliv 244 0.16
(0.35) 0.13
(0.47) 0.13
(0.32) 0.092
(0.70) 0.14 (0.36) 0.14
(0.37) 0.14
Table 5summarizes the results of the comparison. The altimetric SLAs are compared for each
satellite and for the entire time series available by the Root Mean Square Error (RMSE) and the Persons
correlation coefficient. ERS-1 shows a very good correlation for the Honningsvåg tide gauge, where
Remote Sens. 2019,11, 1672 16 of 29
there is ice free year round and a good data coverage; a good correlation for Ny Ålesund, fair correlation
for Vise Ostrov, weak correlation for Sannikova Proliv and no correlation for Golomianyi Ostrov. In the
later tide gauges, we know that the ERS-1 data are very sparse and the tide gauges are placed in
a region where sea-ice is changing from season to season. There is a better correlation in the Summer
data than in the Winter data, where the ocean is ice free. The correlation for ERS-2, Envisat and
CryoSat-2 is excellent for Ny Ålesund and Honningsvåg. For Prudehoe Bay in the Canadian sector
the results are also good, except for Envisat where we get a correlation of only 0.21. This is due to
loss of data in this area in the Envisat period. Data from the Russian sector are generally having
a moderate correlation. There are some very large variation in the tide gauge data (>
0.4 m), which
are not captured by the altimetry. This will be discussed in Section 5.3. In Table 5, the last two columns
represent the comparisons for the time series without ERS-1 (1996–) and the total time series with
ERS-1 (Total). The results improve when the ERS-1 data are ignored.
5. Discussion
In this section the results are examined and evaluated.
5.1. The SLA Record
ERS-1 was the first radar satellite measuring in the Arctic, and the quality of useful data are sparse,
especially in the Beaufort Gyre area. Consequently, several grid cells in this period where empty or
close to and therefore not included in the analysis.
It is difficult to assign the quality of the classification in conventional altimetry. In conventional
altimetry it is not possible to identify the leads with the same accuracy as in SAR or SARIn. A wrong
classification could give a bias with respect to SAR/SARIn. In the transition to SAR/SARIn this could
give negative trends and thus an underestimation of the actual Arctic SLA trend. This concern is partly
supported by the fact that we generally observe smaller sea level trends in the combined Envisat and
CryoSat-2 period compared with the individual Envisat and CryoSat-2 periods and particularly in
regions with seasonal sea-ice cover.
In Figure 3, the mean SLA for each satellite were shown. The ice-edge is visible in the subfigures
to the East of Svalbard. It is a delicate compromise to keep measurements from conventional satellites
in the ice-covered regions or not, as it is impossible to discriminate between reflections from the top of
the ice or from leads, causing the average for these satellites to be too high. The chosen PP threshold
values may be too loose, but it is a trade-off of either removing some of the signal or getting too many
false-positives. We have chosen to go with the second option, and hopefully removing the faulty
values in a strict outlier detection. Despite our careful editing, we still find this to be a problem that
needs further attention. Getting a more strict PP threshold would lead to areas with very low data
coverage, not being able to get a region wide SLA record relaying on data and not only extrapolation.
An other option could be to use more advanced classification schemes and machine learning to get
a better control of the leads in the sea-ice cover similar to [6365].
There is an agreement between the four satellites in Figure 3on positive averaged SLA values
in the region (80–82, 0E–100E) north of Svalbard. This might indicate that the DTU18 MSS used to
reference the average SLA is too low. CryoSat-2 again shows slightly different signal to the other
satellites. This is suspected to be a consequence of the merger of lead data retracked with LARS and
open ocean data retracked with RADS. Any discrepancies/offset between these two data sets might
cause such a mean signal.
The correlation between ERS-1 and ERS-2 (Figure 2) was moderate. This maybe due to the fact that
the ERS-1 satellite’s data coverage is poor in both time and space or it could be that the overlapping
time period is shorter for this comparison. On the other hand it is very reassuring how well the
remaining satellites are matching in the overlapping periods (Figure 2). Especially, it is interesting
how well the ERS-2 and Envisat data are correlating when they are processed with the same retracker.
Errors in the inter-satellite bias estimation would propagate into the final trend estimation. If the
Remote Sens. 2019,11, 1672 17 of 29
inter-satellite bias would be wrong, it would also have shown up in the total time series of the tide
gauge comparison (See Table 5and Figure A2 in Appendix D).
Gridded satellite data are very sensitive to the coverage of satellite tracks. In the Arctic, the data
distribution is much more dense in the high latitudes when using a constant area grid. Furthermore,
the collocation method we are using is very sensitive to missing data and tends to extrapolate data
towards zero. Therefore, using this gridding method where there is no or very little data coverage,
should be done with caution.
5.2. Error Analysis Evaluation
In order to avoid a troublesome error analysis, where the risk of forgetting some error propagation
in the process, a bootstrap analysis was carried out. The advantage with bootstrapping is that we
do not need to know all the error sources and the size of the uncertainties. The concern with block
bootstrapping is to determine the right block size, such that data are independent. The bootstrap
method fails if the blocks are too small and we do not get the right variation of data if the blocks are
too large. This was tested on trail and error.
Data were not normal distributed (Appendix C). This means that a normal standard deviation
evaluation of the uncertainty is not a proper evaluation. In Figure 6, the 95% confidence level was
shown. The uncertainty decreases with time with the largest error in the ERS-1 period and smallest
for the CryoSat-2 measurements. This is not surprising with the large improvements in the satellites
payload giving more data with higher quality.
There is a higher uncertainty in the interior of the Arctic Ocean, than in areas without permanent or
seasonal sea-ice cover (Figure 4). Looking at the standard deviation (not shown here—for comparison
only) of the bootstrapping realizations the level is between about 2 cm in the interior and about 5 cm
outside. This may indicate, that some data are tracked as sea-ice, but the results look similar to Poisson
et al.
(only using Envisat data) which get a transition between the open ocean and the interior of
the Arctic Ocean of about 2 to 4 cm, and much better than the former DTU data set by Cheng et al.
In Figure 7(Section 4.4), it is striking how low the uncertainty in Sector I is compared to the other
sectors. This is the only area with very good satellite coverage, large areas with no sea-ice year round,
and with only a little seasonal sea-ice in the northern parts of the sector. The figures clearly indicate
that conventional altimetry in the interior of the Arctic Ocean (Sector II, III, IV) before 2010 is more
noisy than SAR altimetry from Cryosat-2.
5.3. Regional and Seasonal Variability
In the Arctic Ocean a sea level rise of 1.54 mm/year with a 95% confidence interval of
1.16–1.81 mm/year from September 1991 to September 2018 is found. Ignoring the ERS-1 data,
a linear trend of 2.22 mm/year with a 95% confidence interval of 1.67–2.54 mm/year from January
1996 to September 2018 is found. These results correspond well with other studies:
Cheng et al. [24]
used reprocessed RADS data to make the Arctic DTU SLA record V2 and found a SLA trend of
1.3 mm/year in the period 1993–2011. Andersen and Piccioni
made an update of the
Cheng et al.
data and found a trend of 2.2
1.1 mm/year in the period 1993–2015. Both studies
used data from the sea-ice cover that was processed with an ocean retracker and sampled to 1 Hz.
Prandi et al. [23]
made an update to the SSALTO/DUACS product from 1993–2009 and found a higher
SLA trend of 3.6
1.3 mm/year, but they have a very low data coverage in the area corresponding to
Sector III.
In Figure 9, the SLA trends are derived by cutting the total time series with five years at a time.
Up to 2007 we see an almost constant sea level trend around 1–2 mm/year, but an increased trend
can be seen after 2004. There is also evidence, that the loss of multi-year ice is stable up to 2007 where
after a steepened loss of multi-year ice is seen [
]. Also the Greenland ice sheet did experience an
accelerating ice mass loss already from 2004 [67].
Remote Sens. 2019,11, 1672 18 of 29
Figure 9.
Sea level trends in the CCI DTU/TUM SLA period in steps of eliminating five years at a time.
The light blue color shows the 95% confidence interval.
In general, the tide gauges show slightly higher sea level variability than the altimetry data
(Figures 8and A2). However, in a few of the tide gauges we suspect that the combination of seasonal
sea-ice cover and the location of the tide gauge in sheltered environment away from the harsh Arctic
conditions (i.e., up a river) causes the gauge to measure a signal which is smaller than in the open
ocean. The local GIA signal can be large, and in Table 4, for the Ny Ålesund tide gauge, we saw how
large the difference between the GPS uplift versus the GIA model could be. We also know [
] how
uncertain the GIA model can be, so applying the GIA correction can be associated with large errors.
Sector II was examined in Section 4.4 (Figure 7), Section 4.5 (Figure 8c–e), Appendix D(Figure A2c–e)
and Table 5and found to be the most difficult sector for the altimetry with the worst correlation to the
tide gauges.
Inspecting the seasonal variability for the Arctic Ocean in the different sectors (Figure 7), the mean
SLA for each month (1996–2019) is plotted (Figure 10) for the entire Arctic Ocean (All) and for each of
the geographical sectors (Sector I to IV). We see a maximum SLA for the entire region in October and
a minimum in April. This is similar to [25].
There is a minor difference in the seasonal signal in particularly in Sector IV (Figure 7d) which
can be explained from the fact that conventional altimetry did not have that many observations in
the Canadian Archipelagos, so the seasonal variation was dominated by the seasonal variation in the
Baffin Bay and the Beaufort Gyre. This is not the case with the Cryosat-2 data.
Figure 10.
The mean value for all months in the interval 1996–2019 for the entire Arctic Ocean (blue)
and the different regions outlined in Figure 1: Sector I (orange), II (green), III (red), IV (purple).
The trend pattern in Sector I (Figure 7a) is similar to other studies [
] with exceptions of the
area near the Greenlandic coast, where [
] has a very high SLA trend. For Sector I, we see a maximum
SLA in October and a minimum in April, but as described in Volkov et al.
, the seasonal minimum
and maximum can change within small regions. Volkov et al.
studied the causes for the sea level
variability in this region. The region was divided into smaller areas, and they found a difference in
the maximum amplitude (September to December) depending on the area in question (and minimum
Remote Sens. 2019,11, 1672 19 of 29
in (
March to May
)). The Barents Sea had a phase lag of 1–3 months compared to the Norwegian and
Greenland Sea. This was due to mass changes caused by wind forcing, a varying bottom topography
and dissipation. The annual cycles did not change over different time spans.
Sector II (Figure 7b), the Russian Arctic, has a insignificant trend of 0.04 mm/year with
a confidence interval of
0.86–1.00 mm/year. This is a product of negative trends particular in
the Laptev Sea and positive trends in the Kara Sea and in the East Siberian Sea. In the period before
the satellite era 1954–1989 a trend of 1.85 mm/year was found in the Russian sector from in-situ
measurements [
], where also negative trends in the inner Kara Sea and in the Laptev Sea were
found. 35% of the sea level rise was determined to originate from the ocean expansion, 30% from
the atmospheric pressure, 10% from wind forcing and about 25% from increasing ocean mass from
melting land ice. A more recent study by Henry et al.
covering 1950–2009, also using in-situ
observations agrees on these observations. Furthermore, they studied the GIA effect and found that
the GIA signal is large in this region and models do not agree well. We do not see the large GIA
signal in Figure 7b, but we are looking at a much larger area, where the local effects can vary a lot (see
ex. Table 4). The annual SLA variation (Figure 10, green curve) is highest in November and lowest
in April.
Sector III (Figure 7c), the Beaufort Gyre region, is the sector with the largest trend (5.77 mm/year
with confidence interval 4.12–6.42 mm/year) and with a local maximum trend of 8.45 mm/year.
The large trend is due to increasing fresh water accumulation caused by anti-cyclonic winds and
Ekman transport [
]. There is evidence that the Beaufort Gyre SLA trend has decreased from
1995 to 2003 (
1.3 mm/year) and steepened from 2003 to 2011 (18.8
0.9 mm/year) [
The steepening could also be visible in Figure 7b. Our maximum is smaller, but we are also looking at
a different time interval and a larger geographical region. This region has a maximum annual SLA
peak (Figure 10, red curve) already in September, a second peak in July and a minimum in April. In a
study using moorings [
] from 2003–2007, the authors also find a seasonal cycle with two maximums
in June–July and in November–January. They explain the two maximums as originating from the
largest yearly sea-ice melt and from the largest wind curl when the salt from sea-ice formation has not
jet reached its highest
level. Armitage et al. [25]
have also studied this area, and evaluated the steric
height from 2003–2014. They found a maximum in November, a second peak in June and a minimum
in May. This is shifted compared to this study, but it is seen before [
] that the steric and altimetric
height not necessarily have the same annual cycle. The GIA effect in this area is large (Figure 7b).
Sector IV (Figure 7d), is an area with very different conditions, having both the archipelagos and
the flow through the Baffin Bay. In the archipelagos there are few observations from conventional
altimetry. The main signal comes from the Baffin Bay and the CryoSat-2 satellite. There is a seasonal
SLA peak in January and a low in May (Figure 10, purple). It is also a region with high GIA
values, probable due to melt from the large outlet glaciers in the area. The trend pattern is similar
to Carret et al. [72].
6. Conclusions
The Arctic Ocean is warming faster than ever, nevertheless the Polar Oceans are not included
in global sea level studies due to the uniqueness of the regions and with the associated large errors.
In this study, we have presented the CCI DTU/TUM Arctic SLA record including data from four ESA
radar altimeter satellites, which is (to current date) the longest time series available. We have carefully
combined data from different processings including L2 measurements and state-of-the-art retracking.
This can be troublesome, and a lot of cautions have to be taken in combining different time series. Data
are validated against six tide gauges spread along the coast in the Arctic Ocean. We found a very good
correlation of data in the Fram Strait and a less good correlation in the Russian Arctic due to a bad data
coverage. The GIA estimation is uncertain, but when applying the correction, the sea level rise gets
larger. A sea level rise of 1.54 mm/year with a 95% confidence interval of 1.16–1.81 mm/year is found
in the total time period from September 1991 to September 2018. Ignoring the ERS-1 data and looking
Remote Sens. 2019,11, 1672 20 of 29
at the period from 1996 to 2018, we get a linear trend of 2.22 mm/year with a 95% confidence interval
of 1.67–2.54 mm/year. We handled the troublesome error analysis by a bootstrapping method allowing
us to get an uncertainty estimate without keeping track of all the uncertainties in the processing chain.
The trends are associated with relatively large uncertainties. Our trends are likely underestimated
in the ice-covered regions of the Arctic, which is a combination of several factors currently under
investigation. We had to be tolerant in the editing of the conventional altimetry data in order to get
data at all during the ERS-1/ERS-2/Envisat period. The risk is here, that we allow for reflections from
the top of the ice, biasing the first part of the time series too high. Vice versa the SAR altimetry from
Cryosat-2 over the sea-ice cover is also associated with uncertainties, because we only have SAR data
and not SARIn data. This way we are unable to detect off-nadir reflections, which will cause the sea
level estimate to be too low in the last part of the time series, and hence the estimated trend will be too
low. This could also explain why the altimeteric trend for several stations north of Russia were lower
than that observed at the tide gauge.
In several of our results we do seem to see small effects (i.e., Figure 3) related to the behavior
of the retracker in the presence of partly to full sea-ice coverage. Biases between data sets processed
from different missions and retrackers shall be resolved by cross-calibration, as shown in this study.
For Cryosat-2 the merger between RADS used in the open ocean and LARS (in-house gaussian
retracker) used in the sea-ice seems to result in smaller seasonal SSH effects. The latter effect is
particularly hard to tackle as the data are disjoint to each other. This is currently work in-progress to
improve the consistency among the different retrackers.
Monthly gridded SLA maps from September 1991 to September 2018 is available at https://ftp.sp and at ESA SL_CCI (
Author Contributions:
Conceptualization and methodology: S.K.R., O.B.A., software, formal analysis,
investigation, writing—original draft preparation: S.K.R., validation: S.K.R., O.B.A., C.A.L., data curation:
S.K.R, M.P. (ALES+ retracking), ALES+ data storage and organization: C.S., supervision, funding acquisition:
O.B.A. M.P., writing—review and editing: S.K.R., O.B.A., M.P.
This research was funded by ESA Climate Change Initiative Sea level Budget Closure, contract
No. 4000119910/17/I-NB.
The authors thank Benjamin D. Gutknecht for providing the VLM grids. The first author also
acknowledge Karina Nielsen for fruitful discussions and recommendations in the bootstrap analysis. Furtmore,
the authors are gratefull to the reviewers for constructive and positive review.
Conflicts of Interest: The authors declare no conflict of interest.
The following abbreviations are used in this manuscript:
ALES Adaptive Leading Edge Subwaveform
AO Arctic Oscillation
DAC Dynamic Atmosphere Correction
DTU Technical University of Denmark
ECMWF European Centre for Medium-Range Weather Forecasts
ESA European Space Agency
ERS European Remote Senesing satellite
GIA Glacial Isostatic Ajustment
IBE Inverse Barometer Effect
IPCC Intergovernmental Panel on Climate Change
LRM Low Resolution Mode
MAD MediAn Deviation
MSS Mean Sea Surface
PP Pulse Peakiness
PSMSL Permanent Service for Mean Sea Level
RADS Radar Altimetry Database System
Remote Sens. 2019,11, 1672 21 of 29
REAPER Reprocessing of Altimeter Product for ERS
SSH Sea Surface Height
SAR Synthetic Aperture Radar
SARIn SAR Interferometry
SGDR Sensor Geophysical Data
SLA Sea Level Anomaly
SLBC_CCI Sea Level Budget Closure Climate Change Initiative
SL_CCI Sea Level Climate Change Initiative
TUM Technical University of Munich
Appendix A. Satellite Specific Processing Details
Appendix A.1. ERS-1
ERS-1 was launched in July 1991 carrying among other instruments a pulse-limited single
frequency K
band (13.8 GHz) Radar Altimeter (RA). RA measured with a footprint of 16–20 km
spatial resolution and with an accuracy of 10 cm over the ocean. It was the first Earth observing ESA
satellite with a Sun-synchronous polar orbit (inclination: 98.52
) allowing measurements up to about
. ERS-1 had a repeat cycle of: 3-days, 35-days and 176-days. The mission failed in March 2000,
but already in 1996 the ERS-2 satellite (launched April 1995) took over the operational services [
For ERS-1 we use the Reprocessing of Altimeter Products for ERS (REAPER) [34] L2 data set.
There are known orbital errors for the ERS satellites as a consequence of lacking accuracy of gravity
data and International Terrestrial Reference Frame (ITRF) realizations. It was found necessary to correct
for orbital errors even though one of the REAPER project scopes was to make a better orbit solution [
For ERS-1, we use a orbit correction scheme similar to [
], described in more details
in Cheng et al. [24]
This study is deviating from [
] by not aligning data to the TOPEX/Jason-1/2 SLA because the
coverage of the satellites only reached up to 66
N. To get rid of the most noisy measurements, abnormal
outliers (>10 m) are removed and a median deviation (MAD) outlier detection is applied to each track
before further analysis.
Appendix A.2. ERS-2
ERS-2 was launched in the same orbital plan as ERS-1, but with a one day lag allowing for
tandem measurements. ERS-2 was launched with almost identical instruments as ERS-1 with a few
improvements. ERS-2 was sending valuable measurements back to the ground station till June 2003
and failed entirely in 2011 [
]. For ERS-2 we use the ESA Sensor Geophysical Data Records (SGDR) of
ERS-2 REAPER [34] covering the period from September 1995 to July 2003.
Appendix A.3. Envisat
Envisat is ERS-2’s successor. It was launched in March 2002, as the largest ever built satellite with
10 different instruments aboard. ESA lost contact to Envisat in May 2012. The radar altimeter on-board
(RA-2) was a pulse-limited dual-frequency radar operating in the K
(13.575 GHz) band and S bands.
Only the K
band is used here. The spatial resolution is 2–10 km [
] with an accuracy better than 4.5 cm.
The satellite’s turning latitude was 82
]. Envisat had a repeat cycle of 35-days [
]. The SGDR
Envisat version 2.1 is used. For Envisat the entire duration of the phase 2 (May 2002—October 2010)
and phase 3 (November 2010–May 2012) is used.
The ERS-2 and Envisat satellites are processed with the ALES+ retracker [
]. It is an upgraded
version of the Adaptive Leading Edge Subwaveform (ALES) Retracker [
] that is a retracker
adapted to coastal ocean areas, without lowering the quality of the results in the open ocean.
ALES+ is developed to improve retrievals of peaky waveforms such as echos from leads in the
sea-ice cover. Particularly, one large advantage of this retracker is the seamless transitions between
leads and open ocean waveforms. For unknown reasons data is missing in ERS-2 covering the
15 April 2000
7 May 2000
2 July 2000
9 July 2000
21 January 2001
4 February 2001
Remote Sens. 2019,11, 1672 22 of 29
10 March 2002
17 March 2002
. For Envisat data are missing in the weeks:
16 March 2003
23 March 2003 and 22 March 2011 to 29 March 2011.
Appendix A.4. CryoSat-2
Cryosat-2 was launched on April 2010 and is still active. CryoSat-2 is a dedicated cryosphere
satellite with a coverage up to 88
latitude measuring more of the Arctic than ever before. It has
a 369-day repeat cycle. The main instrument of CryoSat-2 is the K
(13.6 GHz) band SIRAL-2
(SAR/Interferometric Radar Altimeter-2). SIRAL is able to measure in one of three modes; LRM
over ocean and flat surfaces such as the interior of the ice sheets; SAR mode mainly over sea-ice
covered ocean; SARIn over steep terrain and coastal areas. We will be using data from all three modes.
The conventional LRM is a pulse-limited footprint, in SAR mode the Doppler principle results in
a narrow along-track footprint which can be seen as a beam-limited footprint. The SARIn mode utilizes
the two antennas on CryoSat-2 forming an across-track interferometer. The echoes received by each
antenna undergo Doppler beam processing as in SAR mode, but the number of waveforms averaged is
lower due to the longer interval between the bursts. Processing with multi-looks results in a waveform
with a more sharpened leading edge and stronger peak power [
]. For SAR and SARIn baseline C Ice
level 1B data are used.
The CryoSat-2 data contains LRM, SAR and SARIn. For LRM and SAR the 1 hz Radar Altimetry
Database System (RADS) [
] are used. The specular lead returns from the 20 hz SAR and SARIn are
retracked using the Lars Advanced Retracking System (LARS) [
] using a Gaussian fitting routine
similar to [
]. Off-nadir SARIn data are processed as SAR data. For unknown reasons almost one
month of SAR/SARIn data is missing from 12 August 2010 to 16 September 2010.
Appendix B. The Ice Concentration Grid
In each ice concentration (Section 2.2) grid cell a percentage of the ice concentration is given.
A threshold greater than 15% is defined as a cell with sea-ice, and if the cell is below 15% the cell is
classified as open ocean. The ice concentration masks are also used to remove CryoSat-2 RADS data
from the ice cover, such that the RADS data is only used over the open ocean.
Every satellite point is tracked in the sea-ice concentration grid and evaluated in terms of its
location with respect to the sea-ice cover. The satellite coordinate is tracked by a k-dimensional (kd)-tree
for quick nearest-neighbor lookup for the closest coordinate.
Appendix C. SLA Distributions and Uncertainty Estimates
For each monthly data set, 1000 new bootstrap realizations are generated by splitting data in
non-overlapping blocks as done in the making of the SLA (Section 3.6). The bootstrapping blocks need
to be independent from each other, hence the size of the blocks was tested. This test was done as a trial
and error. When the block sizes were too small the bootstrapping failed. Furthermore, it was a wish
to get so much variation in the data as possible, hence the block size should be as small as possible.
The final block sizes are three times the size of the first averaging i.e., 0.6
in the latitude and
longitude direction. The bootstrapping is carried out by randomly drawing
blocks with replacements
from the SLA data set. In practice, this means that some blocks are appearing more times and some
are not represented at all. For each monthly grid cell, for all 1000 bootstrap realizations, a new SLA is
calculated as done in the previous section, and a 95% confidence level is estimated for every grid cell.
The SLAs in each grid cell in the Arctic Ocean is not normal distributed. Figure A1 shows the
monthly SLA distribution of the median of the 1000 bootstrap realizations for selected grid cells
distributed in the Arctic Ocean. It is only valid to show results with a standard deviation if the results
are normal distributed. The Arctic SLA distributions are not normal distributed for all grid cells in the
Arctic Ocean, therefore the uncertainty is expressed in a 95% confidence level. This was also verified
by q-q plots (not shown here). Therefor we use 95% confidence level as the uncertainty estimate.
Remote Sens. 2019,11, 1672 23 of 29
Figure A1.
Examples of median bootstrap SLA distribution from January 1991 to September 2018 for
various grid points. The vertical lines represent the median of the original SLA (red) and the median
bootstrap realizations (blue).
Appendix D. Tide Gauge Comparison
The altimetry data were average with a radius of 350 km around te tide gauge with an inverse
distance weighted function. Various radii around the tide gauges and a simple median average were
tested, but this method showed the best results for the overall result. For all tide gauges, except for Ny
Ålesund, the inverse distance weighting of data (not shown) improves the results. This may indicate
that there is control of the outliers. The reason why Ny Ålesund does not improve by the inverse
distance weighting, could be due to the fact that the tide gauge is situated on an island, where there
are a lot of coastal area with many fjords. To make this point stronger it is also ERS-1 and ERS-2 that
suffer the most when doing the weighted mean. The worse performance of ERS-1 and ERS-2 near the
coast are due to the lower pulse repetition frequency of the satellites. A simple median average of
the data would give a better solution in areas with little data ex. for ERS-1 and in the Russian Arctic.
The amplitude of the altimeter data gets larger for the weighted solution and also the trends get closer.
The tide gauge in the Laptev Sea (Sannikova Proliv) performs best when eliminating data with an
interpolation error >0.10 (not shown). This is normal an indicator of missing or very low data coverage.
In Figure A2 the total altimetry time series (blue) within 350 km of the tide gauge is compared
with the tide gauge data (orange). Also the trend lines are shown and the different trends are written
in the figure caption. The uncertainties are given by three times the standard deviation.
Remote Sens. 2019,11, 1672 24 of 29
Figure A2. Cont.
Remote Sens. 2019,11, 1672 25 of 29
Figure A2.
Tide gauge comparison of the continued altimetry time series from the six tide gauges
in the period from 1991 to 2019, where blue is the altimetry data and orange is the tide gauge data.
The vertical lines are the trend lines colored like the time series. The various trends are written in
the captions. (
) Honningsvåg. (
) Ny Ålesund. (
) Vise Ostrov.(
) Golomianyi Ostrov.(
) Sannikova
Proliv.(f) Prudhoe bay.
Serreze, M.C.; Stroeve, J.C. Arctic sea ice trends, variability and implications for seasonal ice forecasting.
Philos. Trans. R. Soc. A Math. Phys. Eng. Sci. 2015,373, 20140159. [CrossRef] [PubMed]
Church, J.; Clark, P.; Cazenave, A.; Gregory, J.; Jevrejeva, S.; Levermann, A.; Merrifield, M.; Milne, G.;
Nerem, R.; Nunn, P.; et al. Sea level change. In Climate Change 2013: The Physical Science Basis; Contribution
of Working Group I to the Fifth Assessment Report of the Intergovernmental Panel on Climate Change;
PM Cambridge University Press: Cambridge, UK, 2013; pp. 1137–1216. [CrossRef]
Wigley, T.M.L.; Raper, S.C.B. Thermal expansion of sea water associated with global warming. Nature
330, 127–131. [CrossRef]
Cheng, L.; Abraham, J.; Hausfather, Z.; Trenberth, K.E. How fast are the oceans warming? Science
363, 128–129. [CrossRef] [PubMed]
5. Lettenmaier, D.; Milly, P. Land waters and sea level. Nat. Geosci. 2009,2, 452–454. [CrossRef]
Rignot, E.; Velicogna, I.; Van Den Broeke, M.R.; Monaghan, A.; Lenaerts, J. Acceleration of the contribution
of the Greenland and Antarctic ice sheets to sea level rise. Geophys. Res. Lett. 2011,38, 1–5. [CrossRef]
Moon, T.; Joughin, I.; Smith, B.; Howat, I. 21st-century evolution of Greenland outlet glacier velocities.
Science 2012,336, 576–578. [CrossRef] [PubMed]
Yang, Q.; Dixon, T.H.; Myers, P.G.; Bonin, J.; Chambers, D.; van den Broeke, M.R. Recent increases in Arctic
freshwater flux affects Labrador Sea convection and Atlantic overturning circulation. Nat. Commun.
7, 10525. [CrossRef]
Ricker, R.; Hendricks, S.; Beckers, J.F. The impact of geophysical corrections on sea-ice freeboard retrieved
from satellite altimetry. Remote Sens. 2016,8, 317. [CrossRef]
Remote Sens. 2019,11, 1672 26 of 29
Fetterer, F.M.; Drinkwater, M.R.; Jezek, K.C.; Laxon, S.W.; Onstott, R.G. Sea ice altimetry. In Microwave
Remote Sensing of Sea Ice; Carsey, F.D., Ed.; DTIC Document; American Geophysical Union: Washington,
DC, USA, 1992.
Connor, L.N.; Laxon, S.W.; Ridout, A.L.; Krabill, W.B.; McAdoo, D.C. Comparison of Envisat radar and
airborne laser altimeter measurements over Arctic sea ice. Remote Sens. Environ.
,113, 563–570.
Chelton, D.B.; McCabe, P.J. A review of satellite altimeter measurement of sea surface wind speed: With a
proposed new algorithm. J. Geophys. Res. 1985,90, 4707. [CrossRef]
Armitage, T.; Davidson, M. Using the Interferometric Capabilities of the ESA CryoSat-2 Mission to Improve
the Accuracy of Sea Ice Freeboard Retrievals. IEEE Trans. Geosci. Remote Sens.
,52, 529–536. [CrossRef]
Stammer, D.; Ray, R.D.; Andersen, O.B.; Arbic, B.K.; Bosch, W.; Carrère, L.; Cheng, Y.; Chinn, D.S.;
Dushaw, B.D.; Egbert, G.D.; et al. Reviews of Geophysics Accuracy assessment of global barotropic
ocean tide models. Rev. Geophys. 2014,52, 243–282. [CrossRef]
Scharroo, R.; Visser, P. Precise orbit determination and gravity field improvement for the ERS satellites.
J. Geophys. Res. Ocean. 1998,103, 8113–8127. [CrossRef]
Perovich, D.K.; Tucker, W.B.; Eicken, H.; Tucker, W.B.; Perovich, D.K. Indirect measurements of the
mass balance of summer Arctic sea ice with an electromagnetic induction technique. Ann. Glaciol.
33, 194–200.
Henry, O.; Prandi, P.; Llovel, W.; Cazenave, A.; Jevrejeva, S.; Stammer, D.; Meyssignac, B.; Koldunov, N. Tide
gauge-based sea level variations since 1950 along the Norwegian and Russian coasts of the Arctic Ocean:
Contribution of the steric and mass components. J. Geophys. Res. Oceans 2012,117, 1–23. [CrossRef]
18. Rose, S.; Skourup, H.; Forsberg, R. Arctic tides from GPS on sea-ice. J. Geodyn. 2013,63. [CrossRef]
Cazenave, A.; Lombard, A.; Llovel, W. Present-day sea level rise: A synthesis. C. R. Geosci.
,340, 761–770.
Laxon, S.; Peacock, N.; Smith, D. High interannual variability of sea ice thickness in the Arctic region.
Nature 2003,425, 947–950. [CrossRef] [PubMed]
Laxon, S.W.; MacAdoo, D. Arctic ocean gravity field derived from ERS-1 Satellite Altimetry. Science
265, 621–624. [CrossRef]
Peacock, N.R.; Laxon, S.W. Sea surface height determination in the Arctic Ocean from ERS altimetry.
J. Geophys. Res. 2004,109, C07001. [CrossRef]
Prandi, P.; Ablain, M.; Cazenave, A.; Picot, N. A New Estimation of Mean Sea Level in the Arctic Ocean
from Satellite Altimetry. Mar. Geod. 2012,35, 61–81. [CrossRef]
Cheng, Y.; Andersen, O.B.; Knudsen, P. An Improved 20-Year Arctic Ocean Altimetric Sea Level Data Record.
Mar. Geod. 2015,38, 146–162. [CrossRef]
Armitage, T.W.K.; Bacon, S.; Ridout, A.L.; Thomas, S.F.; Aksenov, Y.; Wingham, D.J. Arctic sea surface height
variability and change from satellite radar altimetry and GRACE, 2003–2014. J. Geophys. Res. Oceans
121, 4303–4322. [CrossRef]
Andersen, O.B.; Piccioni, G. Recent Arctic Sea Level Variations from Satellites. Front. Mar. Sci.
,3, 76.
Schaeffer, P.; Faugére, Y.; Legeais, J.F.; Ollivier, A.; Guinle, T.; Picot, N. The CNES_CLS11 Global Mean Sea
Surface Computed from 16 Years of Satellite Altimeter Data. Mar. Geod. 2012,35, 3–19. [CrossRef]
Andersen, O.B.; Knudsen, P. DNSC08 mean sea surface and mean dynamic topography models. J. Geophys.
Res. Oceans 2009,114, 1–12. [CrossRef]
EUMETSAT. EUMETSAT Ocean and Sea Ice Satelitte Application Facility. Global Sea Ice Concentration Continuous
Reprocessing Online Product; Norwegian and Danish Meteorological: Darmstadt, Germany, 2018.
EUMETSAT. EUMETSAT Ocean and Sea Ice Satellite Application Facility. Global Sea Ice Concentration Climate Data
Record 1979–2015 (v2.0, 2017); Norwegian and Danish Meteorological Institutes: Darmstadt, Germany, 2017.
Remote Sens. 2019,11, 1672 27 of 29
Holgate, S.J.; Matthews, A.; Woodworth, P.L.; Rickards, L.J.; Tamisiea, M.E.; Bradshaw, E.; Foden, P.R.;
Gordon, K.M.; Jevrejeva, S.; Pugh, J. New Data Systems and Products at the Permanent Service for Mean Sea
Level. J. Coast. Res. 2013,288, 493–504. [CrossRef]
PSML. Permanent Service for Mean Sea Level, Tide Gauge Data. Available online:
data/obtain (accessed on 20 December 2018).
European Space Agency (ESA); University College London (UCL). CryoSat Product Handbook; Technical
Report; ESRIN-ESA and Mullard Space Science Laboratory—University College London: London, UK, 2013.
Gilbert, L.; Baker, S.; Dolding, C.; Vernier, A.; Brockley, D.; Martinez, B.; Gaudelli, J. REAPER Product
Handbook for ERS Altimetry Reprocessed Products. Available online:
ssed-products (accessed on 20 December 2018).
Passaro, M.; Rose, S.; Andersen, O.; Boergens, E.; Calafat, F.; Dettmering, D.; Benveniste, J. ALES+: Adapting
a homogenous ocean retracker for satellite altimetry to sea ice leads, coastal and inland waters. Remote Sens.
Environ. 2018,211, 456–471. [CrossRef]
European Space Agency (ESA). ENVISAT RA2/MWR Product Handbook; European Space Agency: Paris,
France , 2007.
Stenseng, L.; Andersen, O.B. Preliminary gravity recovery from CryoSat-2 data in the Baffin Bay.
Adv. Space Res. 2012,50, 1158–1163. [CrossRef]
Scharroo, R.; Leuliette, E.; Lillibridge, J.; Byrne, D.; Naeije, M.; Mitchum, G. RADS: Consistent multi-mission
products. In Proceedings of the Symposium on 20 Years of Progress in Radar Altimetry (ESA SP-710),
Venice, Italy, 24–29 September 2012; pp. 5–8.
Dee, D.P.; Uppala, S.M.; Simmons, A.J.; Berrisford, P.; Poli, P.; Kobayashi, S.; Andrae, U.; Balmaseda, M.A.;
Balsamo, G.; Bauer, P.; et al. The ERA-Interim reanalysis : Configuration and performance of the data
assimilation system. Q. J. R. Meteorol. Soc. 2011,137, 553–597. [CrossRef]
Scharroo, R.; Smith, W. A GPS-based climatology for the total electron content in the ionosphere.
J. Geophys. Res. 2010,115. [CrossRef]
Jee, G.; Lee, H. B.; Kim, Y. H.; Chung, J. K.; Cho, J. Assessment of GPS global ionosphere maps (GIM) by
comparison between CODE GIM and TOPEX/Jason TEC data: Ionospheric perspective. J. Geophys. Res.
Space Phys. 2010,115. [CrossRef]
42. Llewellyn, T.S.; Bent, R.B. Documentation and Description of the Bent Ionospheric Model; Atlantic Science Corp
Indian Harbour Beach: Brevard, FL, USA, 1973.
Carrere, L.; Faugère, Y.; Ablain, M. Major improvement of altimetry sea level estimations using
pressure-derived corrections based on ERA-Interim atmospheric reanalysis. Ocean Sci.
,12, 825–842.
Carrère, L.; Lyard, F. Modeling the barotropic response of the global ocean to atmospheric wind and pressure
forcing Comparisons with observations. Geophy. Res. Lett. 2003,30, 4. [CrossRef]
Carrere, L.; Lyard, F.; Cancet, M.; Guillot, A. FES 2014, a new tidal model on the global ocean with enhanced
accuracy in shallow seas and in the Arctic region. In Proceedings of the EGU General Assembly Conference
Abstracts, Vienna, Austria, 12–17 April 2015; Volume 17, p. 5481.
Cartwright, D.E.; Edden, A.C. Corrected tables of tidal harmonics. Geophys. J. Int.
,33, 253–264.
47. Wahr, J.M. Deformation induced by polar motion. J. Geophys. Res. 1985,90, 9363–9368. [CrossRef]
Passaro, M.; Nadzir, Z.A.; Quartly, G.D. Improving the precision of sea level data from satellite altimetry
with high-frequency and regional sea state bias corrections. Remote Sens. Environ.
,218, 245–254.
Andersen, O.B.; Stenseng, L.; Piccioni, G.; Knudsen, P. The DTU15 MSS (Mean Sea Surface) and DTU15LAT
(Lowest Astronomical Tide) reference surface. In Proceedings of the ESA Living Planet Symposium 2016
(SP-740), Prague, Czech Republik, 9–13 May 2016.
Fernandes, M.J.; Lázaro, C.; Ablain, M.; Pires, N. Improved wet path delays for all ESA and reference
altimetric missions. Remote Sens. Environ. 2015,169, 50–74. [CrossRef]
51. Laxon, S.; Rapley, C. Radar altimeter data quality flagging. Adv. Space Res. 1987,7, 315–318. [CrossRef]
Ricker, R.; Hendricks, S.; Helm, V.; Skourup, H.; Davidson, M. Sensitivity of CryoSat-2 Arctic sea-ice
freeboard and thickness on radar-waveform interpretation. Cryosphere 2014,8, 1607–1622. [CrossRef]
Remote Sens. 2019,11, 1672 28 of 29
Knudsen, P.; Andersen, O.B.; Tscherning, C.C. Altimetric gravity anomalies in the Norwegian-Greenland
Sea-Preliminary results from the ERS-1 35 days repeat mission. Geophys. Res. Lett.
,19, 1795–1798.
Laxon, S. Sea ice extent mapping using the ERS-1 radar altimeter. EARSeL Adv. Remote Sens.
,3, 112–116.
Andersen, O.B. Shallow water tides in the northwest European shelf region from TOPEX/POSEIDON
altimetry. J. Geophys. Res. 1999,104, 7729–7741. [CrossRef]
Ablain, M.; Cazenave, A.; Valladeau, G.; Guinehut, S. A new assessment of the error budget of global Mean
Sea Level rate estimated by satellite altimetry over 1993–2008. Ocean Sci. 2009,5, 193–201. [CrossRef]
Lahiri, S. Resampling Methods for Dependent Data; Springer Science & Business Media: Berlin/Heidelberg,
Germany, 2013; p. 374.
Caron, L.; Ivins, E.R.; Larour, E.; Adhikari, S.; Nilsson, J.; Blewitt, G. GIA Model Statistics fro GRACE
Hydrology, Cryosphere, and Ocen Science. Geophy. Res. Lett. 2018,45, 2203–2212. [CrossRef]
Barletta, V.R.; Sabadini, R.; Bordoni, A. Isolating the PGR signal in the GRACE data: Impact on mass balance
estimates in Antarctica and Greenland. Geophys. J. Int. 2008,172, 18–30. [CrossRef]
Barletta, V.R.; Sørensen, L.S.; Forsberg, R. Scatter of mass changes estimates at basin scale for Greenland and
Antarctica. Cryosphere 2013,7, 1411–1432. [CrossRef]
Santamaría-Gómez, A.; Gravelle, M.; Collilieux, X.; Guichard, M.; Míguez, B.M.; Tiphaneau, P.;
Wöppelmann, G. Mitigating the effects of vertical land motion in tide gauge records using a state-of-the-art
GPS velocity field. Glob. Planet. Chang. 2012,98–99, 6–17. [CrossRef]
Sato, T.; Okuno, J.; Hinderer, J.; MacMillan, D.S.; Plag, H.P.; Francis, O.; Falk, R.; Fukuda, Y. A geophysical
interpretation of the secular displacement and gravity rates observed at Ny-Ålesund, Svalbard in the
Arctic—Effects of post-glacial rebound and present-day ice melting. Geophys. J. Int.
,165, 729–743.
Lee, S.; Im, J.; Kim, J.; Kim, M.; Shin, M.; Cheol Kim, H.; Quackenbush, L.J. Arctic Sea Ice Thickness
Estimation from CryoSat-2 Satellite Data Using Machine Learning-Based Lead Detection. Remote Sens.
8, 698. [CrossRef]
Paul, S.; Hendricks, S.; Ricker, R.; Kern, S.; Rinne, E. Empirical parametrization of envisat freeboard retrieval
of arctic and antarctic sea ice based on CryoSat-2: Progress in the ESA climate change initiative. Cryosphere
2018,12, 2437–2460. [CrossRef]
Poisson, J.C.; Quartly, G.D.; Kurekin, A.A.; Thibaut, P.; Hoang, D.; Nencioli, F. Development of an ENVISAT
altimetry processor providing sea level continuity between open ocean and arctic leads. IEEE Trans. Geosci.
Remote Sens. 2018,56, 5299–5319. [CrossRef]
Maslanik, J.; Stroeve, J.; Fowler, C.; Emery, W. Distribution and trends in Arctic sea ice age through
spring 2011. Geophys. Res. Lett. 2011,38, 2–7. [CrossRef]
Velicogna, I.; Wahr, J. Acceleration of Greenland ice mass loss in spring 2004. Nature
,443, 329–331.
[CrossRef] [PubMed]
Volkov, D.L.; Landerer, F.W.; Kirillov, S.A. The genesis of sea level variability in the Barents Sea. Cont. Shelf Res.
2013,66, 92–104. [CrossRef]
Proshutinsky, A.; Ashik, I.M.; Dvorkin, E.N.; Häkkinen, S.; Krishfield, R.A.; Peltier, W.R. Secular sea level
change in the Russian sector of the Arctic Ocean. J. Geophys. Res. Oceans 2004,109, 1–19. [CrossRef]
Proshutinsky, A.; Krishfield, R.; Timmermans, M.L.; Toole, J.; Carmack, E.; McLaughlin, F.; Williams, W.J.;
Zimmermann, S.; Itoh, M.; Shimada, K. Beaufort Gyre freshwater reservoir: State and variability from
observations. J. Geophys. Res. 2009,114, 1–25. [CrossRef]
Giles, K.A.; Laxon, S.W.; Ridout, A.L.; Wingham, D.J.; Bacon, S. Western Arctic Ocean freshwater storage
increased by wind-driven spin-up of the Beaufort Gyre. Nat. Geosci. 2012,5, 194–197. [CrossRef]
Carret, A.; Johannessen, J.A.; Andersen, O.B.; Ablain, M.; Prandi, P.; Blazquez, A.; Cazenave, A. Arctic Sea
Level During the Satellite Altimetry Era. Surv. Geophys. 2017,38, 251–275. [CrossRef]
73. European Space Agency (ESA). ESA Earth Observation Missions; ESA: Paris, France, 2018.
Remote Sens. 2019,11, 1672 29 of 29
Knudsen, P.; Brovelli, M. Collinear and cross-over adjustment of Geosat ERM and Seasat altimeter data in
the Mediterranean Sea. Surv. Geophys. 1993,14, 449–459. [CrossRef]
Passaro, M.; Cipollini, P.; Vignudelli, S.; Quartly, G.D.; Snaith, H.M. ALES: A multi-mission adaptive
subwaveform retracker for coastal and open ocean altimetry. Remote Sens. Environ.
,145, 173–189.
2019 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access
article distributed under the terms and conditions of the Creative Commons Attribution
(CC BY) license (
... Only in recent years have a few basin-wide, multi-annual, gridded datasets of sea surface height been generated at monthly timescales (Armitage et al., 2016;Rose et al., 2019;Prandi et al., 2021). These datasets play an important role in improving our understanding of the Arctic system as a whole and of its present and future change (Timmermans and Marshall, 2020). ...
... However, it is not well known how these products compare to each other or to what extent their spatial and temporal resolution is robust in ice-covered regions (e.g., signal-to-noise ratio). Sea surface height maps have been assessed mostly against tide gauge data at the periphery of the Arctic Ocean or in ice-covered regions against data from hydrographic profiles, which makes it difficult to evaluate the robustness of monthly estimates (Morison et al., 2012;Mizobata et al., 2016;Armitage et al., 2016;Morison et al., 2018;Rose et al., 2019;Morison et al., 2021;Prandi et al., 2021). Furthermore, so far only one study by Armitage et al. (2017) has provided and evaluated monthly maps of geostrophic velocities. ...
... First, to correct ocean tides, we used the FES2014 model (Lyard et al., 2021), a more recent version of the FES2004 model (provided by the ESA as standard correction product; Lyard et al., 2006). FES2014 was previously found to perform better than FES2004 in the Arctic (Cancet et al., 2018) and has already been used to correct the most recent satellite altimetry products in this region (e.g., Rose et al., 2019;Prandi et al., 2021). Furthermore, in support of our choice, we found that the noise on the monthly fields, in areas of high tidal amplitude, was reduced by 20 % by using FES2014 with respect to FES2004 (Appendix A). ...
Full-text available
Satellite altimetry missions flying over the ice-covered Arctic Ocean have opened the possibility of further understanding changes in the ocean beneath the sea ice. This requires complex processing of satellite signals emerging from the sea surface in leads within the sea ice, with efforts to generate consistent Arctic-wide datasets of sea surface height ongoing. The aim of this paper is to provide and assess a novel gridded dataset of sea surface height anomaly and geostrophic velocity, which incorporates both the ice-covered and open ocean areas of the Arctic. Data from the CryoSat-2 mission in the period 2011–2020 were gridded at monthly intervals, up to 88∘ N, using the Data-Interpolating Variational Analysis (DIVA) method. To examine the robustness of our results, we compare our dataset to independent satellite data, mooring time series and Arctic-wide hydrographic observations. We find that our dataset is well correlated with independent satellite data at monthly timescales. Comparisons to in situ ocean observations show that our dataset provides reliable information on the variability of sea surface height and surface geostrophic currents over geographically diverse regions of the Arctic Ocean and different dynamical regimes and sea ice states. At all comparison sites we find agreement with in situ observed variability at seasonal to interannual timescales. Furthermore, we find that our geostrophic velocity fields can resolve the variability of boundary currents wider than about 50 km, a result relevant for studies of Arctic Ocean circulation. Additionally, large-scale seasonal features emerge. Sea surface height exhibits a wintertime Arctic-wide maximum, with the highest amplitude over the shelves. Also, we find a basin-wide seasonal acceleration of Arctic slope currents in winter. We suggest that this dataset can be used to study not only the large-scale sea surface height and circulation, but also the regionally confined boundary currents. The dataset is available in netCDF format from PANGAEA at (Doglioni et al., 2021d).
... Geostrophic velocity at 8° and 10.1°W is referenced to the depth-averaged velocity from the moorings and is interpolate linearly between 11.5° and 8°W. West of 11.5°W where there are no moorings and flow is mostly barotropic, we reference at the surface with a 3-month climatology of surface velocity obtained from two ADT datasets (Andersen & Knudsen, 2009;Rose et al., 2019a) (see Text S1 in Supporting Information S1). ...
... Velocity from both ADT products show similar seasonality with stronger southward velocity in winter than in summer, however, the mean values of the two ADT datasets over the section differ by 4 cm/s (Figure 3e). West of 15°W the difference exceeds 10 cm/s, there the product of shows a strong northward velocity corresponding to the EGCCC persisting year-round, while the product of Rose et al. (2019a) shows general southward velocity that only becomes northward in summer (Figure 4a). Between 15°W and 11.5°W, both products show weak southward velocity year-round. ...
... This relates to the small mean FWT in the EGCCC from the summer 2013-2021 CTD data, which is largely due to the IdF sections, as the 2020-2021 full sections showed a stronger EGCCC (Table 1). Nevertheless, this reversal agrees with the seasonality of surface velocity from the product of Rose et al. (2019a) (Figure 4a) and with the near-surface mooring observations from Münchow et al. (2020), however, it disagrees with the product of and the mooring observations from Topp & Johnson (1997) which showed that the EGCCC flows northward year-round. The EGCCC accounts for 23% of the year-round FWT (Table 1), as there is no further evidence for its southward reversal in winter, our estimate II is possibly too high. ...
Full-text available
We assess the contribution of flow over the Northeast Greenland Shelf (NEGS) to the total freshwater transport (FWT) through the Fram Strait. We analyze available observations since 2013 consisting of hydrographic sections, new mooring data, and surface‐geostrophic velocity and ERA5 winds. We provide first seasonal estimates of the FWT over the entire shelf and find a net southward FWT between 47 and 56 mSv (reference salinity 34.9), or ∼40%–45% of the total FWT through the Fram Strait. In summer, the long‐term mean FWT east of 8°W cumulates to 27 mSv southward, there is no trend toward increasing FWT, and new mooring data show a large seasonal cycle. We show that the NEGS contributes significantly to the Arctic freshwater outflow, however, uncertainty is high. To close the Arctic freshwater budget and provide better FWT estimates to the North Atlantic, sustained year‐round observations on the NEGS are essential.
... Sea ice observed from space has been widely investigated with intensive remote sensing data [8]. Specifically, the sea ice extent, drift, growth stage, concentration, and thickness can be estimated using different sensors, e.g., passive microwave [9,10], scatterometer [11,12], radar altimeter [4,13], and synthetic aperture radar (SAR) [14,15]. However, passive microwave and scatterometer products are typ- 50%, and (c) 92%, respectively. ...
... The aforementioned sea level change patterns have also been reported previously; such asChen et al., 2019; Church et al. 2013a, b; Couldrey et al., 2021; Fox-Kemper et al. 2021;Gregory et al., 2016; Lyu et al., 2020, and common to all the models. As in previous studies(Prandi et al., 2012;Rose et al., 2019; Xiao et al., 2020), the highest sea-level rise is also found in our models in the Arctic Ocean.Many changes in DSL displayed in Figs. 2a, b and c can be associated with changes in the vertically integrated large-scale circulation as depicted by BSF (Figs. 2d e and f). ...
The dependence of future regional sea level changes on ocean model resolution is investigated based on Max Planck Institute Earth System Model (MPI-ESM) simulations with varying spatial resolution, ranging from low resolution (LR), high resolution (HR), to eddy-rich (ER) resolution. Each run was driven by the shared socioeconomic pathway (SSP) 5-8.5 (fossil-fueled development) forcing. For each run the dynamic sea level (DSL) changes are evaluated by comparing the time mean of the SSP5-8.5 climate change scenario for the years 2080–99 to the time mean of the historical simulation for the years 1995–2014. Respective results indicate that each run reproduces previously identified large-scale DSL change patterns. However, substantial sensitivity of the projected DSL changes can be found on a regional to local scale with respect to model resolution. In comparison to models with parameterized eddies (HR and LR), enhanced sea level changes are found in the North Atlantic subtropical region, the Kuroshio region, and the Arctic Ocean in the model version capturing mesoscale processes (ER). Smaller yet still significant sea level changes can be found in the Southern Ocean and the North Atlantic subpolar region. These sea level changes are associated with changes in the regional circulation. Our study suggests that low-resolution sea level projections should be interpreted with care in regions where major differences are revealed here, particularly in eddy active regions such as the Kuroshio, Antarctic Circumpolar Current, Gulf Stream, and East Australian Current. Significance Statement Sea level change is expected to be more realistic when mesoscale processes are explicitly resolved in climate models. However, century-long simulations with eddy-resolving models are computationally expensive. Therefore, current sea level projections are based on climate models in which ocean eddies are parameterized. The representation of sea level by these models considerably differs from actual observations, particularly in the eddy-rich regions such as the Southern Ocean and the western boundary currents, implying erroneous ocean circulation that affects the sea level projections. Taking this into account, we review the sea level change pattern in a climate model with featuring an eddy-rich ocean model and compare the results to state-of-the-art coarser-resolution versions of the same model. We found substantial DSL differences in the global ocean between the different resolutions. Relatively small-scale ocean eddies can hence have profound large-scale effects on the projected sea level which may affect our understanding of future sea level change as well as the planning of future investments to adapt to climate change around the world.
... With the advance in space technology, altimetry observations have played a significant role in Earth science (Guo et al. 2022a), such as the monitoring of sea level variation (Ablain et al. 2015;Rose et al. 2019;Watson et al. 2015), investigation on ocean geology and plate tectonics (Hwang and Chang 2014;Sandwell and Smith 2009;Sandwell et al. 2014) inversion of marine gravity anomaly (Andersen et al. 2010;Annan and Wan, 2021;Hwang 1998;Sandwell et al. 2013Sandwell et al. , 2021 and seafloor topography Hu et al. 2021;Smith and Sandwell 1994;Tozer et al. 2019). ...
Full-text available
In this study, China’s first altimeter satellite Haiyang-2A (HY-2A) data combined observations from CryoSat-2, SARAL/AltiKa, and Jason-1&2 are used to calculate the global (60°S–60°N) marine deflections of the vertical and gravity anomalies named Global Marine Gravity Anomaly Version 1(GMGA1), with grid resolution of 1′ × 1′. The deflections of the vertical from each satellite observations are first derived from the gradients of the geoid height through the least squares method. The deflections of the vertical are then merged by assigning different weights to each satellite product based on their accuracy. Finally, gravity anomalies are obtained by the remove-restore method. The results reveal that the fused deflections of the vertical have an accuracy of 0.4 arcsec in the north component and 0.8 arcsec in the east component. HY-2A’s contribution to the north component of the integrated deflections of the vertical is second only to Cryosat-2. Jason-1/2 accounts for a large proportion of the integrated east components. Compared to worldwide products such as DTU17, Sandwell & Smith V31.1, as well as values from EGM2008, EIGEN-6C4 and XGM2019e_2159, GMGA1 has an accuracy of around 3.3 mGal. By not using HY-2A data, the precision of GMGA1 is reduced by about 0.1 mGal. To further improve the accuracy, seafloor topography information is used to provide short wavelength gravity anomaly. It is verified in the South China Sea (112°E–119E°, 12°N–20°N) using the Parker formula. By combining shipborne depth generated data and GMGA1 through a filtering technique, a new version of gravity anomaly grid with an accuracy improvement of 0.4 mGal in the South China Sea is obtained. Graphical Abstract
... However, due to the remote location of the Arctic and its relatively harsh environmental conditions, the availability of observational input is limited. For instance, information on the Arctic Ocean sea surface height (SSH) is needed for many purposes; from studying the influence of Arctic glacier melt on the regional sea level (e.g., Cazenave et al. 2019;Rose et al. 2019) to monitoring sea ice thickness (e.g., Laxon et al. 2013;Wernecke and Kaleschke 2015). Unfortunately, in situ data are limited to a few tide gauges at the coast and the presence of sea ice hampers measurements by satellite altimeters. ...
Full-text available
In the Arctic Ocean, obtaining water levels from satellite altimetry is hampered by the presence of sea ice. Hence, water level retrieval requires accurate detection of fractures in the sea ice (leads). This paper describes a thorough assessment of various surface type classification methods, including a thresholding method, nine supervised-, and two unsupervised machine learning methods, applied to Sentinel-3 Synthetic Aperture Radar Altimeter data. For the first time, the simultaneously sensed images from the Ocean and Land Color Instrument, onboard Sentinel-3, were used for training and validation of the classifiers. This product allows to identify leads that are at least 300 meters wide. Applied to data from winter months, the supervised Adaptive Boosting, Artificial Neural Network, Naïve-Bayes, and Linear Discriminant classifiers showed robust results with overall accuracies of up to 92%. The unsupervised Kmedoids classifier produced excellent results with accuracies up to 92.74% and is an attractive classifier when ground truth data is limited. All classifiers perform poorly on summer data, rendering surface classifications that are solely based on altimetry data from summer months unsuitable. Finally, the Adaptive Boosting, Artificial Neural Network, and Bootstrap Aggregation classifiers obtain the highest accuracies when the altimetry observations include measurements from the open ocean. © 2022 The Author(s). Published by Informa UK Limited, trading as Taylor & Francis Group.
Sea‐level change in the Arctic Ocean (AO) is a key indicator of the rapidly changing Arctic climate. Changes in steric mass as well as atmospheric and oceanic circulation patterns contribute significantly to AO sea‐level variability. Monitoring of AO sea level is not as simple as in the other global oceans due to several factors such as ice cover, and limitation associated with retrieval of satellite measurements in the polar regions, etc. Recent developments in retracking of satellite altimeter data, and improved algorithms have enabled nearly pan‐Arctic observation of AO sea‐level changes. In this chapter, we summarize recent findings using remotely‐sensed sea‐level estimates in the AO and further assess the representation of the same in the new‐generation climate models participating in Coupled Model Intercomparison Project‐6 (CMIP6).
Full-text available
Sea-level variations in coastal areas can differ significantly from those in the nearby open ocean. Monitoring coastal sea-level variations is therefore crucial to understand how climate variability can affect the densely populated coastal regions of the globe. In this paper, we study the sea-level variability along the coast of Norway by means of in situ records, satellite altimetry data, and a network of eight hydrographic stations over a period spanning 16 years (from 2003 to 2018). At first, we evaluate the performance of the ALES-reprocessed coastal altimetry dataset (1 Hz posting rate) by comparing it with the sea-level anomaly from tide gauges over a range of timescales, which include the long-term trend, the annual cycle, and the detrended and deseasoned sea-level anomaly. We find that coastal altimetry and conventional altimetry products perform similarly along the Norwegian coast. However, the agreement with tide gauges in terms of trends is on average 6 % better when we use the ALES coastal altimetry data. We later assess the steric contribution to the sea level along the Norwegian coast. While longer time series are necessary to evaluate the steric contribution to the sea-level trends, we find that the sea-level annual cycle is more affected by variations in temperature than in salinity and that both temperature and salinity give a comparable contribution to the detrended and deseasoned sea-level variability along the entire Norwegian coast. A conclusion from our study is that coastal regions poorly covered by tide gauges can benefit from our satellite-based approach to study and monitor sea-level change and variability.
Full-text available
Satellite altimetry plays a key role in monitoring changes in sea level and climate change. The quality of satellite altimetry products is commonly ensured through dedicated calibration. One such calibration is with microwave transponders acting as ground reference point targets. It is common practice that satellite ranges between the transponder phase center and the satellite center of gravity (CoG) are compared against the true geometric ranges to determine bias. Transponder ranges are, however, realized by the two phase centers of the altimeter and the ground transponder. So, to make this comparison feasible, the space origin of the measured range is transferred from the altimeter phase center (APC) to the satellite CoG by applying a constant offset, usually referred to as “CoG correction”. Instead of a fixed “CoG correction”, this work introduces the actual vector between APC and CoG in space, by examining the satellite attitude. Thus, the observed and geometric distances to the transponder are both referred to the APC. The case of Jason-3 and Sentinel-6A Michael Freilich (Sentinel-6A MF) with two transponders on Crete (CDN1) and Gavdos (GVD1) islands is examined. At first, the attitude of Jason-3 is determined by its quaternions. Then, analysis reveals that the transponder bias is correlated with the Jason-3 satellite attitude. The revised calibration brings about bias changes which fluctuate from about −2 mm to 1 mm in range and from −110μs to +110 μs in datation for Jason-3. Spectral analysis on the bias differences between the revised and conventional transponder calibrations reveals constituents with periods of 117, 39 and 23 days. Finally, the revised methodology on crossover calibrations over the GVD1 transponder results in an improvement between the mean bias of the ascending and descending orbits by 12% for Jason-3 and by 14% (preliminary) for Sentinel-6A MF.
Due to the existence of seasonal or perennial sea ice cover, the determination of the Arctic sea surface is more difficult than that of mid-low latitudinal oceans. Focusing on the sea surface height in the ice-covered region, this paper constructs a new Arctic mean sea surface (MSS) model, named SUST22, by combining the measurements from ICESat and Cryosat-2 missions. The lead detection methods of ICESat and Cryosat-2 are first studied and modified to acquire sea surface measurements with better accuracy. The results have shown that the standard deviation of Cryosat-2-derived Arctic sea surface height is about 3–4 cm in 10-km resolution grids, while the value of ICESat is 5–6 cm. Then the MSS construction procedure is discussed and the SUST22 MSS model is constructed. The new model is compared with the other four Arctic MSS models. The best agreement is found between SUST22 and DTU21 with an average difference of −4.0 ± 5.2 cm. These models are also validated by ICESat-2 samples. The average difference between ICESat-2 and SUST22 is 15.8 ± 7.4 cm, which shows that the new model SUST22 presents better consistency with the ICESat-2 than any of the other models.
Full-text available
Observational records of ocean heat content show that ocean warming is accelerating
Full-text available
The sea state bias (SSB) is a large source of uncertainty in the estimation of sea level from satellite altimetry. It is still unclear to what extent it depends on errors in parameter estimations (numerical source) or to the wave physics (physical source). By improving the application of this correction we compute 20-Hz sea level anomalies that are about 30% more precise (i.e. less noisy) than the current standards. The improvement is two-fold: first we prove that the SSB correction should be applied directly to the 20-Hz data (12 to 19% noise decrease); secondly, we show that by recomputing a regional SSB model (based on the 20-Hz estimations) even a simple parametric relation is sufficient to further improve the correction (further 15 to 19% noise decrease). We test our methodology using range, wave height and wind speed estimated with two retrackers applied to Jason-1 waveform data: the MLE4 retracked-data available in the Sensor Geophysical Data Records of the mission and the ALES retracked-data available in the OpenADB repository ( The regional SSB models are computed parametrically by means of a crossover analysis in the Mediterranean Sea and North Sea. Correcting the high-rate data for the SSB reduces the correlation between retracked parameters. Regional variations in the proposed models might be due to differences in wave climate and remaining sea-state dependent residual errors. The variations in the empirical model with respect to the retracker used recall the need for a specific SSB correction for any retracker. This study, while providing a significantly more precise solution to exploit high-rate sea level data, calls for a re-thinking of the SSB correction in both its physical and numerical component, gives robustness to previous theories and provides an immediate improvement for the application of satellite altimetry in the regions of study.
Full-text available
In order to derive long-term changes in sea-ice volume, a multi-decadal sea-ice thickness record is required. CryoSat-2 has showcased the potential of radar altimetry for sea-ice mass-balance estimation over the recent years. However, precursor altimetry missions such as Environmental Satellite (Envisat) have not been exploited to the same extent so far. Combining both missions to acquire a decadal sea-ice volume data set requires a method to overcome the discrepancies due to different footprint sizes from either pulse-limited or beam-sharpened radar echoes. In this study, we implemented an inter-mission-consistent surface-type classification scheme for both hemispheres, based on the waveform pulse peakiness, leading-edge width, and surface backscatter. In order to achieve a consistent retracking procedure, we adapted the threshold first-maximum retracker algorithm, previously used only for CryoSat-2, to develop an adaptive retracker threshold that depends on waveform characteristics. With our method, we produce a global and consistent freeboard data set for CryoSat-2 and Envisat. This novel data set features a maximum monthly difference in the mission-overlap period of 2.2 cm (2.7 cm) for the Arctic (Antarctic) based on all gridded values with spatial resolution of 25 km × 25 km and 50 km × 50 km for the Arctic and Antarctic, respectively.
Full-text available
Water level from sea ice-covered oceans is particularly challenging to retrieve with satellite radar altimeters due to the different shapes assumed by the returned signal compared with the standard open ocean waveforms. Valid measurements are scarce in large areas of the Arctic and Antarctic Oceans, because sea level can only be estimated in the openings in the sea ice (leads and polynyas). Similar signal-related problems affect also measurements in coastal and inland waters. This study presents a fitting (also called retracking) strategy (ALES+) based on a subwaveform retracker that is able to adapt the fitting of the signal depending on the sea state and on the slope of its trailing edge. The algorithm modifies the existing Adaptive Leading Edge Subwaveform retracker originally designed for coastal waters, and is applied to Envisat and ERS-2 missions. The validation in a test area of the Arctic Ocean demonstrates that the presented strategy is more precise than the dedicated ocean and sea ice retrackers available in the mission products. It decreases the retracking open ocean noise by over 1 cm with respect to the standard ocean retracker and is more precise by over 1 cm with respect to the standard sea ice retracker used for fitting specular echoes. Compared to an existing open ocean altimetry dataset, the presented strategy increases the number of sea level retrievals in the sea ice-covered area and the correlation with a local tide gauge. Further tests against in-situ data show that also the quality of coastal retrievals increases compared to the standard ocean product in the last 6 km within the coast. ALES+ improves the sea level determination at high latitudes and is adapted to fit reflections from any water surface. If used in the open ocean and in the coastal zone, it improves the current official products based on ocean retrackers. First results in the inland waters show that the correlation between water heights from ALES+ and from in-situ measurement is always over 0.95.
Full-text available
Over the Arctic regions, current conventional altimetry products suffer from a lack of coverage or from degraded performance due to the inadequacy of the standard processing applied in the ground segments. This paper presents a set of dedicated algorithms able to process consistently returns from open ocean and from sea-ice leads in the Arctic Ocean (detection of water surfaces and derivation of water levels using returns from these surfaces). This processing extends the area over which a precise sea level can be computed. In the frame of the European Space Agency Sea Level Climate Change Initiative (, we have first developed a new surface identification method combining two complementary solutions, one using a multiple-criteria approach (in particular the backscattering coefficient and the peakiness coefficient of the waveforms) and one based on a supervised neural network approach. Then, a new physical model has been developed (modified from the Brown model to include anisotropy in the scattering from calm protected water surfaces) and has been implemented in a maximum likelihood estimation retracker. This allows us to process both sea-ice lead waveforms (characterized by their peaky shapes) and ocean waveforms (more diffuse returns), guaranteeing, by construction, continuity between open ocean and ice-covered regions. This new processing has been used to produce maps of Arctic sea level anomaly from 18-Hz ENVIronment SATellite/RA-2 data.
Full-text available
We provide a new analysis of glacial isostatic adjustment (GIA) with the goal of assembling the model uncertainty statistics required for rigorously extracting trends in surface mass from the Gravity Recovery and Climate Experiment (GRACE) mission. Such statistics are essential for deciphering sea level, ocean mass, and hydrological changes because the latter signals can be relatively small (≤2 mm/yr water height equivalent) over very large regions, such as major ocean basins and watersheds. With abundant new >7 year continuous measurements of vertical land motion (VLM) reported by Global Positioning System stations on bedrock and new relative sea level records, our new statistical evaluation of GIA uncertainties incorporates Bayesian methodologies. A unique aspect of the method is that both the ice history and 1-D Earth structure vary through a total of 128,000 forward models. We find that best fit models poorly capture the statistical inferences needed to correctly invert for lower mantle viscosity and that GIA uncertainty exceeds the uncertainty ascribed to trends from 14 years of GRACE data in polar regions.
Full-text available
Results of the sea-level budget in the high latitudes (up to 80°N) and the Arctic Ocean during the satellite altimetry era. We investigate the closure of the sea-level budget since 2002 using two altimetry sea-level datasets based on the Envisat waveform retracking: temperature and salinity data from the ORAP5 reanalysis, and Gravity Recovery And Climate Experiment (GRACE) space gravimetry data to estimate the steric and mass components. Regional sea-level trends seen in the altimetry map, in particular over the Beaufort Gyre and along the eastern coast of Greenland, are of halosteric origin. However, in terms of regional average over the region ranging from 66°N to 80°N, the steric component contributes little to the observed sea-level trend, suggesting a dominant mass contribution in the Arctic region. This is confirmed by GRACE-based ocean mass time series that agree well with the altimetry-based sea-level time series. Direct estimate of the mass component is not possible prior to GRACE. Thus, we estimated the mass contribution from the difference between the altimetry-based sea level and the steric component. We also investigate the coastal sea level with tide gauge records. Twenty coupled climate models from the CMIP5 project are also used. The models lead us to the same conclusions concerning the halosteric origin of the trend patterns.
Full-text available
Satellite altimeters have been used to monitor Arctic sea ice thickness since the early 2000s. In order to estimate sea ice thickness from satellite altimeter data, leads (i.e., cracks between ice floes) should first be identified for the calculation of sea ice freeboard. In this study, we proposed novel approaches for lead detection using two machine learning algorithms: decision trees and random forest. CryoSat-2 satellite data collected in March and April of 2011?2014 over the Arctic region were used to extract waveform parameters that show the characteristics of leads, ice floes and ocean, including stack standard deviation, stack skewness, stack kurtosis, pulse peakiness and backscatter sigma-0. The parameters were used to identify leads in the machine learning models. Results show that the proposed approaches, with overall accuracy >90%, produced much better performance than existing lead detection methods based on simple thresholding approaches. Sea ice thickness estimated based on the machine learning-detected leads was compared to the averaged Airborne Electromagnetic (AEM)-bird data collected over two days during the CryoSat Validation experiment (CryoVex) field campaign in April 2011. This comparison showed that the proposed machine learning methods had better performance (up to r = 0.83 and Root Mean Square Error (RMSE) = 0.29 m) compared to thickness estimation based on existing lead detection methods (RMSE = 0.86?0.93 m). Sea ice thickness based on the machine learning approaches showed a consistent decline from 2011?2013 and rebounded in 2014.
Full-text available
The new dynamic atmospheric correction (DAC) and dry tropospheric (DT) correction derived from the ERA-Interim meteorological reanalysis have been computed for the 1992-2013 altimeter period. Using these new corrections significantly improves sea level estimations for short temporal signals (<ĝ€2 months); the impact is stronger if considering old altimeter missions (ERS-1, ERS-2, and Topex/Poseidon), for which DAC-ERA (DAC derived from ERA-Interim meteorological reanalysis) allows reduction of the along-track altimeter sea surface height (SSH) error by more than 3ĝ€cm in the Southern Ocean and in some shallow water regions. The impact of DT-ERA (DT derived from ERA-Interim meteorological reanalysis) is also significant in the southern high latitudes for these missions. Concerning more recent missions (Jason-1, Jason-2, and Envisat), results are very similar between ERA-Interim and ECMWF-based corrections: on average for the global ocean, the operational DAC becomes slightly better than DAC-ERA only from the year 2006, likely due to the switch of the operational forcing to a higher spatial resolution. At regional scale, both DACs are similar in the deep ocean but DAC-ERA raises the residual crossovers' variance in some shallow water regions, indicating a slight degradation in the most recent years of the study. In the second decade of altimetry, unexpectedly DT-ERA still gives better results compared to the operational DT. Concerning climate signals, both DAC-ERA and DT-ERA have a low impact on global mean sea level rise (MSL) trends, but they can have a strong impact on long-term regional trends' estimation, up to several millimeters per year locally.
Arctic sea surface height (SSH) is poorly observed by radar altimeters due to the poor coverage of the polar oceans provided by conventional altimeter missions and because large areas are perpetually covered by sea ice, requiring specialized data processing. We utilize SSH estimates from both the ice-covered and ice-free ocean to present monthly estimates of Arctic Dynamic Ocean Topography (DOT) from radar altimetry south of 81.5°N and combine this with GRACE ocean mass to estimate steric height. Our SSH and steric height estimates show good agreement with tide gauge records and geopotential height derived from Ice-Tethered Profilers. The large seasonal cycle of Arctic SSH (amplitude ∼5 cm) is dominated by seasonal steric height variation associated with seasonal freshwater fluxes, and peaks in October-November. Overall, the annual mean steric height increased by 2.2±1.4 cm between 2003 and 2012 before falling to circa 2003 levels between 2012 and 2014 due to large reductions on the Siberian shelf seas. The total secular change in SSH between 2003 and 2014 is then dominated by a 2.1±0.7 cm increase in ocean mass. We estimate that by 2010, the Beaufort Gyre had accumulated 4600 km3 of freshwater relative to the 2003-2006 mean. Doming of Arctic DOT in the Beaufort Sea is revealed by Empirical Orthogonal Function analysis to be concurrent with regional reductions in the Siberian Arctic. We estimate that the Siberian shelf seas lost ∼180 km3 of freshwater between 2003 and 2014, associated with an increase in annual mean salinity of 0.15 psu yr-1. Finally, ocean storage flux estimates from altimetry agree well with high-resolution model results, demonstrating the potential for altimetry to elucidate the Arctic hydrological cycle.