QUALITY INFORMATION DOCUMENT For OSI TAC Sea Ice products , -002, -003, -004, -006, -007, -009, -010, -011, -012

Transcription

1 QUALITY INFORMATION DOCUMENT For OSI TAC Sea Ice products , -002, -003, -004, -006, -007, -009, -010, -011, -012 Issue: 2.5 Contributors: Steinar Eastwood (MET Norway), Juha Karvonen (FMI), Frode Dinessen (MET Norway), Andrew Fleming (BAS), Leif Toudal Pedersen (DMI), Roberto Saldo (DTU), Jørgen Buus-Hinkler (DMI), Bruce Hackett (MET Norway), Fanny Ardhuin (IFREMER), Matilde Brandt Kreiner (DMI) Approval Date by Quality Assurance Review Group :

2 Change Record Issue Date Description of Change Author Validated By All Creation of the document All L-A Breivik All Ingested in new template, udated for V3 L-A Breivik L-A Breivik All Corrections after V3 acceptance L-A Breivik All IV Introduced overview dataset for SEAICE_ARC_SEAICE_L4_NRT_ OBSERVATIONS_011_003. Reinstatement of SEAICE_ARC_SEAICE_L4_NRT_ OBSERVATIONS_011_005. Revisions from QuARG comments IV Introduced Arctic sea ice coverage indicator ARC_SEAICE_INDEX_ II, IV Revised system description. Updates for 011_001 (add Antarctic ice drift dataset), 011_007 (quality parameters) II, IV Added info for sea ice drift time series 011_ All Added Appendix with quality info for Arctic ice chart products 002, 003, 004. Ice surface temp product 008 moved to separate QUID document. ARC_SEAICE_INDEX_002 moved to separate QUID document I.3, II Added table in chapter I.3 and a product description in chapter II 2.0 1/9/2015 all Change format to fit CMEMS graphical rules Remove info about SEA ICE product which is retired from catalogue. Introduce references to CMEMS IV.1 Update of val description for OSI SAF NRT product 011_001 L. T. Pedersen, T. Hamre, B. Hackett, F. Dinessen L-A Breivik B. Hackett, T. Lavergne, J. Buus-Hinkler F. Ardhuin, B. Hackett M. B. Jensen, B. Hackett L-A Breivik B. Hackett L-A Breivik L-A Breivik L.-A. Breivik F. Dinessen B. Hackett L. Crosnier, B. Hackett B. Hackett II.1.5, Time extension of OSI SAF REP B. Hackett L. Crosnier EU Copernicus Marine Service Public Page 2/ 58

3 IV.8 product 011_ All Revisions due to review by Mercator Océan I.2.6, I.3, IV.6.1, IV I.2.10, II.7, IV.9 Mosaic dataset added to iceberg concentration 011_007 S. Eastwood, J. Karvonen, J. Buus-Hinkler, M. B. Kreiner B. Hackett EAN/validation information J. Buus-Hinkler B. Hackett Update of input data type. A Fleming C. Wettre EU Copernicus Marine Service Public Page 3/ 58

4 Table of contents I Executive Summary... 6 I.1 Products covered by this document... 6 I.2 Summary of results... 7 I.2.1 Global sea ice L4 - concentration, edge, type, drift... 7 I.2.2 Arctic sea ice L4 Svalbard... 7 I.2.3 Arctic sea ice L4 Greenland... 7 I.2.4 Baltic sea ice L I.2.5 Global high-resolution SAR sea ice drift... 8 I.2.6 Arctic SAR iceberg concentration... 9 I.2.7 Arctic sea ice surface temperature... 9 I.2.8 Global ice concentration time series... 9 I.2.9 Arctic ice drift time series I.2.10 Antarctic high-resolution ice edge I.3 Estimated Accuracy Numbers II Production Subsystem description II.1 Global sea ice L4 - concentration, edge, type, drift II.2 Sea Ice charts from the national ice services at MET Norway, DMI and FMI II.3 Global high-resolution SAR sea ice drift II.4 Arctic SAR iceberg concentration II.5 Global ice concentration time series II.6 Arctic ice drift time series II.7 Antarctic high-resolution ice edge III Validation framework IV Validation IV.1 Global sea ice products IV.1.1 Validation Procedures IV.1.2 Validation results IV.2 Regional Sea Ice Svalbard IV.2.1 Validation procedure IV.2.2 Validation results IV.3 Regional Sea Ice Greenland IV.3.1 Validation procedure IV.3.2 Validation results EU Copernicus Marine Service Public Page 4/ 58

5 IV.4 Regional Sea Ice Baltic IV.4.1 Validation Procedure IV.4.2 Validation results IV.5 Arctic and Antarctic ice drift IV.5.1 Validation procedure IV.5.2 Validation results IV.6 Arctic Ice berg density IV.6.1 Validation procedure: IV.6.2 Validation results IV.7 Global ice concentration time series IV.7.1 Validation procedure IV.7.2 Validation results IV.8 Arctic ice drift time series IV.8.1 Validation procedure: IV.8.2 Validation results IV.9 Antarctic high-resolution ice edge IV.9.1 Validation procedure: IV.9.2 Validation results V Appendix V.1 Ice Chart Uncertainty Estimates V.1.1 Comparison of ice charts and OSI SAF ice concentration dataset in the Arctic V.1.2 Comparison of ice charts in the Baltic Sea V.1.3 Concentration Estimation Exercise at Ice Analyst Workshop V.1.4 Discussion and conclusion EU Copernicus Marine Service Public Page 5/ 58

6 I Executive Summary The main sea ice focus in the CMEMS OSI TAC is on SAR-based products. The Copernicus satellite Sentinel-1, launched April 3, 2014, is developed for operational SAR applications, i.e., in order to provide a stable and frequent access to SAR data for operational users. Providing input satellite data for sea ice mapping is one of the primary goals of the Sentinel-1 mission. In the MyOcean1 project, the ASAR on ENVISAT was the main SAR data source until failure of the satellite. Thereafter, the data stream was replaced by data from the Canadian RADARSAT2. At the time of writing (September 2015), Sentinel-1 SAR data have been implemented, either in replacement of or in addition to RADARSAT2. High-resolution ice charts based on SAR and other satellite data are routinely produced by the national ice services in Denmark, Finland and Norway. In MyOcean1 the link between the ice services and operational numerical oceanography represented by the MFCs was established. Ice charts are transformed to grids and formats usable for the numerical models, and made available through the OSI TAC. These products are used by the MFCs (Global, Arctic and Baltic) for validation of models and for sea ice data assimilation. They are considered being close to the true state of the sea ice. A main task in the MyOcean2 and the MyOcean Follow-On projects was to describe the uncertainties, in terms of possible biases, variance and true resolution of the ice charts. In addition, the OSI TAC redistributes global NRT and reprocessed sea ice products from the EUMETSAT Ocean and Sea Ice Satellite Application Facility (OSI SAF) under an agreement between CMEMS and EUMETSAT. Methods and estimates of the quality of the sea ice products delivered by the OSI TAC are described in the subsequent sections. I.1 Products covered by this document This document provides information to CMEMS users regarding the quality of OSI TAC products delivered through the Dissemination Unit. The product names are listed in the table below. Excluded from this list is the Arctic sea ice surface temperature, SEAICE_ARC_SEAICE_L4_NRT_OBSERVATIONS_011_008, which is described in a separate document. Product Product description Production unit, PU Dissemination unit DU SEAICE_GLO_SEAICE_L4_NRT _OBSERVATIONS_011_001 Global sea ice L4 sea ice concentration, edge, type, drift (OSI SAF product) SIW-METNO-OSLO-NO SIW-METNO-OSLO- NO SEAICE_ARC_SEAICE_L4_NRT _OBSERVATIONS_011_002 Arctic sea ice L4 Svalbard SIW-METNO-OSLO-NO SIW-METNO-OSLO- NO SEAICE_ARC_SEAICE_L4_NRT _OBSERVATIONS_011_003 Arctic sea ice L4 Greenland SIW-DMI- COPENHAGEN-DK SIW-METNO-OSLO- NO SEAICE_BAL_SEAICE_L4_NRT _OBSERVATIONS_011_004 SEAICE_BAL_SEAICE_L4_NRT _OBSERVATIONS_011_011 Baltic sea ice L4 SIW-FMI-HELSINKI-FI SIW-METNO-OSLO- NO Baltic sea ice SAR SIW-FMI-HELSINKI-FI SIW-METNO-OSLO- NO SEAICE_GLO_SEAICE_L4_NRT _OBSERVATIONS_011_006 Global high-resolution SAR sea ice drift SIW-DTUSPACE- COPENHAGEN-DK SIW-METNO-OSLO- NO SEAICE_ARC_SEAICE_L4_NRT _OBSERVATIONS_011_007 Arctic SAR sea iceberg concentration SIW-DMI- COPENHAGEN-DK SIW-METNO-OSLO- NO SEAICE_GLO_SEAICE_REP_O BSERVATIONS_011_009 Global sea ice concentration time series (OSI SAF product) SIW-METNO-OSLO-NO SIW-METNO-OSLO- NO EU Copernicus Marine Service Public Page 6/ 58

7 SEAICE_ARC_SEAICE_REP_O BSERVATIONS_011_010 Arctic sea ice drift time series SIW-IFREMER-BREST- FR SIW-METNO-OSLO- NO SEAICE_ANT_SEAICE_L4_NRT _OBSERVATIONS_011_012 Antarctic high-resolution ice product SIW-BAS-CAMBRIDGE- UK SIW-METNO-OSLO- NO Table 1: OSI TAC Sea Ice products and partner roles. I.2 Summary of results In contrast to other satellite measurements, such as SST and wind, there is a significant lack of useful in situ observations with which to compare the satellite data. This is true of all the variables provided in the sea ice products: concentration, ice edge, ice type, ice drift, iceberg concentration, ice surface temperature. The in situ observations that are available are spotty and intermittent in time. It is therefore very difficult to calculate sound statistics for the global datasets and nearly impossible for the regional datasets. For this reason, a common validation method for the different products has not been established and validation is not following conventions laid out in the CMEMS Cal/Val guidelines. In the following section we have made a short summary of the validation performed for the individual products. A more detailed description can be found in the validation chapter. I.2.1 Global sea ice L4 - concentration, edge, type, drift (SEAICE_GLO_SEAICE_L4_NRT_OBSERVATIONS_011_001) This is the OSI SAF global NRT sea ice product, redistributed by CMEMS OSI TAC. Concentration: For the weekly validation the concentration product is required to have a mean annual bias and standard deviation less than 10 % ice concentration in the Northern Hemisphere. For the Southern Hemisphere the concentration product is required to have a bias and standard deviation less that 15%. These requirements are met with good margins especially during winter. The ice concentration has a negative bias between 1 and 5 % during winter. In the Arctic from May to August, the bias increases to around 10% mainly due to ambiguous open water due to melt ponds on the ice. Edge: The ice edge validations shows that the mean difference to the ice edge in Arctic winter is around 10 km. In summer this can increase to above 20 km. Drift: The validation of ice drift product in the Arctic reveals an annual mean standard deviation of 2-3 km in both components. I.2.2 Arctic sea ice L4 Svalbard (SEAICE_ARC_SEAICE_L4_NRT_OBSERVATIONS_011_002) The sea ice concentration ice charts from the national Norwegian Ice Service are generated based on a manual interpretation of high-resolution satellite data. The main sensor used is data from the SAR sensor on-board Radarsat-2 and Sentinel-1. Due to its high spatial resolution and weather independency SAR data are often considered being close to the true state of the sea ice. The limited amount of in-situ measurements available in the Arctic makes it very difficult to give a quantitative number of the accuracy for this product. In the validation performed in the first quarter of 2013 we have used some optical satellite data from the NOAA satellites and an alternative SAR source from the COSMO SkyMed satellite. The data have been visual compared against the ice charts. A report on estimating the ice chart uncertainty from comparison analyses of different ice chart data sets is attached as appendix to this document. I.2.3 Arctic sea ice L4 Greenland (SEAICE_ARC_SEAICE_L4_NRT_OBSERVATIONS_011_003) The concentration ice charts from the Danish national ice service is produced in 2 flavours: the normal ice chart, which typically covers an approximately 500x500 kilometre subset of the area EU Copernicus Marine Service Public Page 7/ 58

8 observed by a radar satellite, and the overview ice chart, which is produced twice per week and covers the entire area. The nominal quality of the charts is similar and validation procedures are the same. The products are compared against the OSI SAF ice concentration product (SEAICE_GLO_SEAICE_L4_NRT_OBSERVATIONS_011_001) and the differences between the two products are reported; 90-95% of ice edge locations are the same. In case of differences, experience shows that it is more frequent that the ice chart edge shows more ice than the OSI SAF concentration product. Between 80 and 85% of grid points have differences less than 10% and more than 90% of points have differences less than 20%. Standard deviation of differences between the ice charts and the OSI SAF concentration are between 7 and 9% for the analysed period. Biases are in the order of 5% with the ice chart overestimating relative to the OSI SAF product. A report on estimating the ice chart uncertainty from comparison analyses of different ice chart data sets is attached as appendix to this document. I.2.4 Baltic sea ice L4 (SEAICE_BAL_SEAICE_L4_NRT_OBSERVATIONS_011_004 SEAICE_BAL_SEAICE_L4_NRT_OBSERVATIONS_011_011 ) The ice thickness was validated against the available ice thickness measurements (made on Finnish ice breakers), and the agreement was good, especially taking into account that ice thickness can vary significantly even within a small spatial area, and the measurements are point measurements. We have also included a comparison of the previous winter EM ice thickness measurements, because their processing only got ready after the final validation report of the previous season. The ice concentration were compared with the radiometer product of University of Bremen, and taking into account the restrictions of the radiometer product (poor/moderate resolution, mixed pixels), the agreement was good. The ice drift was validated against ice drifter buoys deployed into the Gulf of Bothnia, and also compared to daily ice motion computed from daily COSMO-SkyMED WR mode data (100x100km size, 30m resolution) over a fixed area in the Gulf of Bothnia. Because the ice cover in the drift ice area of the Gulf of Bothnia was not very thick during the season, the ice drift estimation was not very reliable, and the results of the comparisons did not give very good agreement either. This fact was also indicated by the quality index related to each ice drift estimate. The assigned quality estimates for the season were relatively low, indicating probable inaccuracies in the estimates. A report on estimating the ice chart uncertainty from comparison analyses of different ice chart data sets is attached as appendix to this document. I.2.5 Global high-resolution SAR sea ice drift (SEAICE_GLO_SEAICE_L4_NRT_OBSERVATIONS_011_006) The high-resolution SAR sea ice drift product has been compared to high quality positions from in-situ drifters (GPS equipped Ice Tethered Profilers ITPs). Validation has only been carried out in the Arctic as there is no high quality in-situ data from the Antarctic at the moment. There is nothing that indicates that the test results should not also apply (be indicative for) to the Antarctic. The results show very good correlation between the reference and product values (almost 1.0). Mean error values are very small which indicates almost no bias and the standard deviation of the differences is low. Comparison with Ice Tethered Profiler 37 shows a bit higher STD on the error but nothing that would indicate any problems. It is worth noting that the numbers are consistently good also for the smaller sample sizes. EU Copernicus Marine Service Public Page 8/ 58

9 The drift product has been enhanced in V4 by including information from the automated monthly product validation in the product files. I.2.6 Arctic SAR iceberg concentration (SEAICE_ARC_SEAICE_L4_NRT_OBSERVATIONS_011_007) The iceberg number density product is based on satellite data from SAR (Sentinel-1 and Radarsat-2). The output is iceberg number density given as number of icebergs sampled in grid cells each covering 10 x 10 km. The DMI iceberg detection algorithm utilizes a Constant False Alarm Rate (CFAR) concept (see Gill, 2001 and e.g. Basically this means that if a given pixel s backscatter value deviates significantly from the background noise calculated for nearby pixels, then this pixel will be classified as an iceberg-pixel. The main source of error in the product is related to noise in SAR data (speckle, sea-clutter, high winds etc.), which in some cases may lead to the detection of too many (false) iceberg-targets. Due to lack of ground truth data there are currently no statistics on missed targets (i.e. small icebergs not visible to the sensor due to their physical size or due to severe noise in the imagery).thus, uncertainties of the product are derived by comparing the extreme results against about 10 years of iceberg statistics for the Greenland Waters. Error statistics are calculated assuming that all observations with number of icebergs per grid-cell larger than the 97 percentile (about 2xSTD, assuming a normal distribution) may be potential errors. Results show that when errors occur, then typically a few false iceberg-targets appear in the data, however in very rare cases severe noise phenomena may lead to a huge amount of false targets appearing. The data files include an index of extremity and distribution percentiles. These offer the user more information on how to interpret the iceberg density variable. Mosaic dataset: Until the recent upgrade of the product (CMEMS Release V2 April 2016) an output file has been produced for each available SAR scene. The V2 upgrade is a new dataset added to the already existing single-scene dataset. The new dataset consists of mosaics (of gridded iceberg numberdensities). A daily mosaic is produced, which the latest four days of observations and is based solely on Sentinel-1 data. It is important to note that the mosaicked dataset uses a new scheme for sea ice masking, which is based on high resolution sea ice charts instead of passive microwave data (e.g., AMSR-2). The scheme has been set up such that the sea ice chart with the least time difference to the individual satellite observation is used for ice masking. Note that the new mosaicked dataset currently does not include statistical layers in the netcdf files. For this dataset the implementation of these layers is planned to be done when the Sentinel archives include multiple years, and when the new sea ice scheme has been run on existing data through reanalysis I.2.7 Arctic sea ice surface temperature (SEAICE_ARC_SEAICE_L4_NRT_OBSERVATIONS_011_008) A separate QuID document is available for this product. Ref: MYOF-OSI-QUID I.2.8 Global ice concentration time series (SEAICE_GLO_SEAICE_REP_OBSERVATIONS_011_009) This is the OSI SAF global reprocessed sea ice concentration time series, redistributed by CMEMS OSI TAC. The OSI SAF global sea ice concentration reprocessed products are compared to the National Ice Center (NIC, USA) ice charts. EU Copernicus Marine Service Public Page 9/ 58

10 There are three requirements on accuracy: 1) threshold accuracy 20 % (yearly average), 2) target accuracy 10% for the NH-product and 15% for SH-product (yearly average), and 3) optimal accuracy: 10% (yearly average). All requirements on accuracy are met in the comparison with ice charts. The deviations between the ice product and the ice charts are large during summer melt up to 20% on both hemispheres, while during the winter the deviations are 5-10%. It is clear that the NIC ice charts do not necessarily represent the truth, but rather a fairly independent dataset for comparison. I.2.9 Arctic ice drift time series (SEAICE_ARC_SEAICE_REP_OBSERVATIONS_011_010) The low resolution Arctic sea ice drift products have been compared to buoys data (US IABP). The validation shows good correlation between the reference and the product and standard deviation is 7.5 km for 3 day lags drift and 8.9 km for 6 day lag drifts for both velocity components. I.2.10 Antarctic high-resolution ice edge (SEAICE_ANT_SEAICE_L4_NRT_OBSERVATIONS_011_012) The Antarctic high-resolution regional sea ice edge product is an operational product from BAS that is delivered to CMEMS. The product is based on manual interpretation of satellite SAR data from Radarsat-2 ScanSAR Wide and Sentinel-1 Extra Wide Mode SAR imagery. The temporal coverage is variable and focused on the austral summer season (October to March). The sea ice product s spatial extent depends on data availability. Validation of the product is difficult due to the lack of in situ observations and coincident highresolution imagery for comparison. As an alternative the product is compared with a manually interpreted sea ice edge from MODIS and other visible satellite imagery and the OSI SAF sea ice edge product. In general the agreement is very good and normally greater than 90%. The distance between ice edges varies between 1 and 11 km. I.3 Estimated Accuracy Numbers Product Name Type Description SEAICE_GLO_SEAICE_L 4_NRT_OBSERVATIONS _011_001 SEAICE_ARC_SEAICE_L 4_NRT_OBSERVATIONS _011_002 SEAICE_ARC_SEAICE_L 4_NRT_OBSERVATIONS _011_003 SEAICE_BAL_SEAICE_L 4_NRT_OBSERVATIONS _011_004 SEAICE_GLO_SEAICE_L 4_NRT_OBSERVATIONS _011_006 SEAICE_ARC_SEAICE_L 4_NRT_OBSERVATIONS concentration Summer: Bias between 1-5 % edge drift concentration concentration concentration Ice drift Iceberg Winter: Bias ~10% Summer: 20 km Winter: 10 km Standard deviation 2-3 km Only a visual validation against optical satellite data from NOAA AVHRR and MODIS. Generally a good agreement Compared against OSIAF ice concentration: Bias ~5%, STD: 7-9% Compared against radiometer product of University of Bremen: Difference ± 10% Compared against drifters: Very good correlation between the reference and product values ~ 1.0. Potential error defined as positive deviation from EU Copernicus Marine Service Public Page 10/ 58

11 _011_007 density 97 percentile (>~2xSTD) in 10 years of obs.: SEAICE_GLO_SEAICE_ REP_OBSERVATIONS_0 11_009 SEAICE_ARC_SEAICE_ REP_OBSERVATIONS_0 11_010 SEAICE_BAL_SEAICE_L 4_NRT_OBSERVATIONS _011_011 SEAICE_ANT_SEAICE_L 4_NRT_OBSERVATIONS _011_012 Ice Concentratio n Ice drift Ice thickness Ice drift Ice edge Mean potential error (# icebergs): 4. Typical error [P50 of pot. errors] (# icebergs): 2 Minimum error (# icebergs): 0 Largest 1% of errors [0.03% of the observations] (error range: # icebergs): Compared against NIC ice charts Summer: ~20% Winter: ~5-10% Compared against buoys: STD: 7.5 km for 3 day lags STD: 8.9 km for 6 day lags Compared against in-situ measurements: Difference 0-45cm Compared against buoys: Differences 40-80% Compared against OSI SAF sea ice and MODIS: Agreement better than 90% II Production Subsystem description The OSI TAC is a multi-mission integration activity. The main tasks to fulfil are: 1. provide real-time, delayed mode update operations for L3, and L4 data products for SST, Sea Ice and Wind; 2. provide data for long-term sea ice and surface wind monitoring data (climatology); 3. provide quality control, validation and error characterization of data products and services. This document describes the Sea Ice part of the OSI TAC. Wind and SST are described in separate documents. Seven Production Units are processing and delivering sea ice products, and performing the routine quality monitoring and validation of their products. For several of the sea ice products that are delivered, satellite data from SAR are used as the main source of input data due to its high spatial resolution and weather independency. During the MyOcean projects the sea ice partners have established a very good working relationship with ESA in order to ensure access to SAR data, and this cooperation continues in CMEMS. In the preparation to the Sentinel-1a there have been a close discussions regarding the geographical coverage and requested beam modes. Through these discussions the partners have been able to ensure the best possible access for their services. A direct link to the ESA data-hub is established and data from Sentinel-1 are retrieved in NRT by automatically polling for new data files on the ftp server. After a new file has been detected it is transmitted and processed by at the local production unit. The process is similar as for RADARSAT. EU Copernicus Marine Service Public Page 11/ 58

12 II.1 Global sea ice L4 - concentration, edge, type, drift (SEAICE_GLO_SEAICE_L4_NRT_OBSERVATIONS_011_001) MET Norway produces a NRT global sea ice product (concentration, edge, type, drift) and a multi-year time series (concentration) under the auspices of the EUMETSAT Ocean and Sea Ice Satellite Application Facility. These products are produced independently of CMEMS and are delivered through CMEMS by agreement between EUMETSAT and CMEMS The sea ice products have been developed to be derived from passive microwave (SSMIS) and active microwave (ASCAT) sensors. The sea ice concentration product uses SSMIS data, while the sea ice edge and type using multi sensor methods with a Bayesian approach to combine SSMIS and ASCAT data. Concentration: A multi sensor analysis scheme for sea ice concentration analysis has been developed based on a Bayesian approach. The analysis is a 2-step procedure. In the first step ice concentration is calculated in the swath projection for each satellite passage. SSMIS ice concentration is calculated using a combination of the Bristol algorithm and the Bootstrap frequency mode algorithm. In the second step, the multi pass analysis, these results are analyzed on the 10 km OSI SAF grid. Several SSMIS observation nodes, with estimated concentrations, influence on each analysis grid point. The radius of influence, r, for each SSMIS observation is 18 km. Edge: The OSI SAF ice edge product is using the three SSMIS parameters, PR(19), GR(19,37) and PR(91) and the ASCAT parameter anisfmb. In the first step ice class (closed ice, open ice and water) -probabilities are estimated on the satellite swath projection for each passage. For SSMIS the low resolution data PR(19), GR(19,37) are combined. In the second step the ice class probabilities for each parameter is estimated on the OSI SAF 10 km grid on one day of data. The result is three probability estimates that is combined in a Bayesian approach. Type: The OSI SAF ice type product is using the SSMIS parameter GR(19,37) and the ASCAT backscatter parameter. In the first step ice class (first year ice and multi year ice) probabilities are estimated on the satellite swath projection for each passage, in all areas where the ice edge product from the same day shows there is ice. In the second step the ice class probabilities for each parameter is estimated on the OSI SAF 10 km grid on one day of data. The result is two probability estimates that is combined in a Bayesian approach. Drift: Low resolution ice drift datasets are computed on a daily basis from aggregated maps of passive microwave (e.g. SSMIS, SSM/I, AMSR-E) or scatterometer (e.g. ASCAT) signals. The typical resolution/spacing of those input images is 12.5 km. Wide swaths, high repetition rates and independence with respect to the atmospheric perturbations permit daily coverage of most of the sea ice covered regions. In summer, surface melting and a denser atmosphere preclude from the retrieval of meaningful information. During fall, winter and spring, however, the excellent coverage makes it possible to extract 48 hours global ice drift vectors at a spatial resolution of 62.5 km. The algorithm implemented in the OSI SAF chain is based on the CMCC (Continuous Maximum Cross Correlation) method, where pixel values in the sub-images are interpolated from those in the nominal pixels. At this stage, a simple bi-linear interpolation is implemented. This strategy allows the formulation of the image matching problem in a continuous formalism and thus strongly limits the quantization noise. II.2 Sea Ice charts from the national ice services at MET Norway, DMI and FMI (SEAICE_ARC_SEAICE_L4_NRT_OBSERVATIONS_011_002) (SEAICE_ARC_SEAICE_L4_NRT_OBSERVATIONS_011_003) (SEAICE_BAL_SEAICE_L4_NRT_OBSERVATIONS_011_004) The sea ice concentration ice charts from the national Ice service are generated based on a manual interpretation of high-resolution satellite data. Satellite data from Radarsat-2 and Sentinel-1 are EU Copernicus Marine Service Public Page 12/ 58

13 automatically downloaded from the ftp servers at ESA-data hub. Radarsat-2 data are provided in ScanSAR Wide mode dual polarization (HH/HV). One image covers 500x500 km and have a spatial horizontal resolution of 50x50 meter. Sentinel-1 is provided in Extended Wide mode dual polarization (HH/HV). One image covers 400x400km and have a spatial resolution of 20x40 meter. In addition to these SAR data, the analyses are supplemented with optical satellite data from MODIS and NOAA AVHRR. The data are made available for the operator in NRT and are manually analysed with respect to sea ice concentration in a GIS production system. The final analyses are gridded to a 1km netcdf product. II.3 Global high-resolution SAR sea ice drift (SEAICE_GLO_SEAICE_L4_NRT_OBSERVATIONS_011_006) The high-resolution SAR sea ice drift product is produced by a Maximum Cross Correlation (MCC) processing of overlapping SAR data. All SAR data available from the ESA-data hub are automatically downloaded in NRT. The data are resampled to a 300m resolution before the displacement is computed in the MCC algorithm. The nominal temporal span between processed swaths is 24 hours, the nominal product grid resolution is a 10km. The algorithm is supporting SAR data from Envisat ASAR, Radarsat-2 and Sentinel-1. II.4 Arctic SAR iceberg concentration (SEAICE_ARC_SEAICE_L4_NRT_OBSERVATIONS_011_007) The iceberg number density product is based on target (iceberg) detection in data inferred from satellite borne Synthetic Aperture Radar (SAR). Sentinel-1 and Radarsat-2 data are downloaded from the ESA data hub and resampled to 80m and 100m spatial resolution, respectively. The iceberg detection algorithm utilizes a Constant False Alarm Rate (CFAR) concept. Basically this means that if a given pixel (backscatter value) deviates significantly from the background noise calculated for nearby pixels, this pixel will be classified as an iceberg-pixel. To produce the final CMEMS product all iceberg clusters are counted in 10 km grid cells to produce an iceberg number density grid. II.5 Global ice concentration time series (SEAICE_GLO_SEAICE_REP_OBSERVATIONS_011_009) The OSI SAF global sea ice concentration reprocessed products are a reprocessing of the SEAICE_GLO_SEAICE_L4_NRT_OBSERVATIONS_011_001. This product is produced by the EUMETSAT OSI SAF and disseminated by MET Norway. It contains reprocessed sea ice concentration (0-100%) from to II.6 Arctic ice drift time series (SEAICE_ARC_SEAICE_REP_OBSERVATIONS_011_010) The low resolution Arctic sea ice drift products have been compared to buoys data (US IABP). The validation shows good correlation between the reference and the product and standard deviation is 7.5 km for 3 day lags drift and 8.9 km for 6 day lag drifts for both components. II.7 Antarctic high-resolution ice edge (SEAICE_ANT_SEAICE_L4_NRT_OBSERVATIONS_011_012) The Antarctic high-resolution regional sea ice edge product is a manually interpreted product based on satellite data from Radarsat-2 ScanSAR Wide and Sentinel-1 Extra Wide Mode. The satellite data are automatically downloaded from the ftp servers at ESA-data hub and loaded into an ESRI ArcGIS production system where it is analysed with respect to the ice edge. A gridded sea ice concentration EU Copernicus Marine Service Public Page 13/ 58

14 product is available irregularly depending on ship traffic and data availability. The temporal coverage is variable and restricted on the austral summer season (October to March). The three National Ice Services at MET Norway, DMI and FMI are producing NRT high-resolution ice charts mainly based on SAR data for Arctic, Greenland Sea and Baltic. These services are independent of CMEMS. As a part of the OSI TAC service these ice charts are re-projected, gridded and supplied with uncertainty estimates as required for operational use in coupled ocean-sea ice models the Marine Core Service. The Dissemination Unit at MET Norway is distributing all sea ice products, sharing the infrastructure with Arctic MFC. Status of these products is continuously monitored and can be found at: III Validation framework Validation is a continuous on-going activity to characterize accuracy and quality of the delivered sea ice products. It is mainly based on operational data, but can be supported by campaign data. Each PU is responsible for validation of their products. The OSI TAC Sea Ice Validation activities are based on what is already implemented at the partners institutes and has shown to be useful. Validation of the OSI TAC sea ice products is a challenge due to lack of ground truth. The analyses produced by the National Ice Services at MET Norway, DMI and FMI are manually constructed ice charts based on high-resolution satellite data, SAR and IR/optical for the Arctic Ocean, Greenland Sea and Baltic Sea. In general, these products are considered as the best available sea ice information and used as ground truth in several studies of sea ice. The automatic generated global sea ice products are using these manual analyses as reference data. In the Baltic the situation is better as in situ measurements from ships and ice breakers are regularly available through the ice season. Given the variability in the nature of the OSI TAC Sea ice products in terms of the way they are produced and limited availability of in situ reference data, the validation procedures are not standardized. Due to this the OSI TAC sea ice validation are currently not able to follow the standard conventions laid out in the CMEMS Cal/Val guidelines. The validation procedures for each product are described in the following chapters together with short summary of the validation results. IV Validation IV.1 Global sea ice products Product Product description Production unit, PU Dissemination unit DU SEAICE_GLO_SEAICE_L4_NRT_ OBSERVATIONS_011_001 Global sea ice L4 -sea ice cons, edge, type, drift SIW-METNO-OSLO-NO SIW-METNO-OSLO- NO These are global sea ice concentration, edge, type and drift datasets that are operational products from EUMETSAT Satellite Application Facility for Ocean & Sea Ice (OSI SAF). They are delivered to CMEMS for redistribution in order to ease and enhance their use in operational oceanography. Validation of the data is carried out as an OSI SAF activity and is documented in EU Copernicus Marine Service Public Page 14/ 58

15 1. Validation Reports issued for product updates, which are maintained at The latest versions for the current Issue of this document are: ice edge/type: ice drift: ice conc: 2. Monthly updated validation metrics, found at The following sections present some recent results of the validation activities. For more detailed information, the reader is directed to the web addresses above. IV.1.1 Validation Procedures Sea Ice Concentration is validated against 2D gridded data sets (operational Ice Service data from DMI and MET Norway). The results are given in terms of error estimates: STD, and bias of ice concentration. Sea Ice edge is validated against 2D gridded data sets (operational Ice Service data from DMI and MET Norway). The results are given in terms of mean distance to the ice edge and in percentage of overestimation and underestimation. Sea ice type is monitored by looking at the stability/consistency of the area covered by multi year ice. The standard deviation of the difference between daily and running 11-days mean of area covered by multi year ice is monitored. : The OSI SAF ice edge product is using the three SSMIS parameters, PR(19), GR(19,37) and PR(91) and the ASCAT parameter anisfmb. In the first step ice class (closed ice, open ice and water) probabilities are estimated on the satellite swath projection for each passage. For SSMIS the low resolution data PR(19), GR(19,37) are combined. In the second step the ice class probabilities for each parameter is estimated on the OSI SAF 10 km grid on one day of data. The result is three probability estimates that is combined in a Bayesian approach. Validation of ice drift in the Arctic is performed against in situ drifters. The results are given in terms of error estimates: STD, and bias of the x and y components of the drift vector. IV.1.2 Validation results The OSI SAF sea ice products are validated operationally by the OSI SAF. Results from the validation of ice concentration, ice edge and ice drift are listed in the tables below. EU Copernicus Marine Service Public Page 15/ 58

16 Figure 1: Area for comparison of OSI SAF ice concentration and edge with MET Norway Ice Service charts, presented in Table 1 and Table 2. Shown is the Barents Sea with Novaya Semlya at right and Svalbard at left. Table 1: OSI SAF Sea Ice Concentration product (= SEAICE_GLO_SEAICE_L4_NRT_OBSERVATIONS_011_001 / concentration dataset) compared with MET Norway Ice Service charts (= SEAICE_ARC_SEAICE_L4_NRT_- OBSERVATIONS_011_002). Column 2 (3) shows percentage of 10km x 10km pixels with values within ±10% (±20%) of the collocated ice chart value. Bias and Std dev refer to the same collocated values. EU Copernicus Marine Service Public Page 16/ 58

17 Table 2: OSI SAF Sea Ice Edge product product (= SEAICE_GLO_SEAICE_L4_NRT_OBSERVATIONS_011_001 / edge dataset) compared with MET Norway Ice Service charts (= SEAICE_ARC_SEAICE_L4_NRT_- OBSERVATIONS_011_002). Match means ice chart edge within 10kmx10km pixel of OSI SAF edge. OSI SAF underestimates (overestimates) means OSI SAF edge is farther north (south) than ice service edge. In the table below, monthly validation statistics for the multi-sensor/merged sea ice drift product are reported upon. X and Y denote the 2 components of the drift vectors. b() is the bias and s() the standard deviation of the error e(x) = Xproduct - Xreference (in km). Columns 6, 7 and 8 are respectively the slope, the intercept of the regression line, and the Pearson correlation coefficient between Product and Reference data pairs. N is the number of collocation data pairs. EU Copernicus Marine Service Public Page 17/ 58

18 EU Copernicus Marine Service Public Page 18/ 58

19 Table 3: Validation results for the multi-sensor ice drift product in the Northern Hemisphere. Comparison (reference) data are drifting buoys and Envisat SAR (motion vectors); for details see IV.2 Regional Sea Ice Svalbard Product Product description Production unit, PU Dissemination unit DU SEAICE_ARC_SEAICE_L4_NRT_ OBSERVATIONS_011_002 Arctic sea ice L4 Svalbard SIW-METNO-OSLO-NO SIW-METNO-OSLO- NO The regional sea ice product Svalbard is provided by the Sea Ice Service in Tromsø, VNN (MET Norway). It covers European Arctic wit focus on the areas around Svalbard and the Barents Sea. The ice concentration is given wit h1 km horizontal resolution. The ice charts are primarily based on SAR (Radarsat-2) data, together with AVHRR and MODIS data. A detailed interpretation of satellite imagery and a subsequent mapping procedure are carried out by skilled ice analysts. An important use of the products is for validation of ocean/ice models (Arctic and Global MFC) and the global OSI TAC products. IV.2.1 Validation procedure Validation of the products is a challenge due to lack of ground truth. The routine comparison with OSI SAF global ice concentration and ice edge data allows for more regular data quality assurance. The product is at irregular intervals manual compared against high-resolution optical satellite data from the MODIS and NOAA satellites. The validation is focus on situation where weather conditions allow use of optical data. In addition some alternative SAR data from the COSMO-SkyMed satellites have been used for validation. Validation is not carried out in the dark session due to lack of daylight and no visual data available. IV.2.2 Validation results The validation of the regional high-resolution ice chart of the Arctic has been performed in a two step routine. First the accuracy of the product has been validated against alternative satellite sources. Secondly the produced netcdf file has been checked to be compliant withe the CF-1.4 version of the netcdf standard. The figure below shows example of a visual validation against a NOAA image and against a COSMO- SkyMed image. EU Copernicus Marine Service Public Page 19/ 58

20 Figure 2: Visual validation of the regional high-resolution ice chart against a NOAA AVHRR image, northwest of Svalbard, 8 February Figure 3: Visual comparison of COSMO-SkyMed data (left) and ice chart (right) on 10 December A netcdf file has been downloaded from the dissemination unit at MET Norway and the content has been checked against the CF-checker that was developed at the Hadley Centre for Climate Prediction and Research, UK Met Office by Rosalyn Hatcher. The output from this test indicated no errors and no warnings. The latest validation report is found at: IV.3 Regional Sea Ice Greenland Product Product description Production unit, PU Dissemination unit DU SEAICE_ARC_SEAICE_L4_NRT_ OBSERVATIONS_011_003 Arctic sea ice L4 Greenland SIW-DMI- COPENHAGEN-DK SIW-METNO-OSLO- NO The regional Sea Ice products Greenland cover Greenland Waters and are produced and delivered by DMI. These are high-resolution ice concentration with grid resolution of 1 kilometers. The product comes in 2 flavours. The normal ice chart which typically covers an approximately 500x500 kilometer subset of the area as observed by a radar satellite, and the overview ice chart which is produced twice per week and covers the entire area. The nominal quality of the charts is similar and validation procedures are the same. The products (ice charts) are based on manual interpretation of high-resolution satellite data (RADARSAT, TerraSAR-X, Cosmo Skymed, ENVISAT SAR (no longer in operation), MODIS, NOAA and METOP AVHRR, AMSR-E and AMSR2, SSMI and SSMIS,...). Charts are originally produced as vector charts using a GIS and a large number of satellite data sources. The vector charts (polygons of areas with similar ice conditions) are subsequently gridded on a 1 km grid and converted to NetCDF for CMEMS distribution and use. Boundaries of polygons are drawn with a nominal accuracy of approximately 1 Km. The main use of the data in CMEMS is for validation of ocean/ice models (Arctic and Global MFC) and the global SIW TAC products. EU Copernicus Marine Service Public Page 20/ 58

21 IV.3.1 Validation procedure Validation of the high-resolution regional ice chart of the Arctic Ocean/Greenland Waters is a challenging task due to lack of ground truth. In the production the satellite data used is expected to have a resolution high enough to represent the ground truth of the mapped area. The interpretation of the satellite data is a subjective analysis by the operator on duty. To verify how this will influence on the final product we will at regular intervals have two ice analysts produce independent products based on the same input data. The two products will be compared on a pixel basis and a confusion matrix containing ice concentration classes from the two analysts will be generated (V4). This will give a measure of the uncertainty of the different ice concentration classes and the overall uncertainty of the product. We also continuously compare the regional charts with the OSI SAF global sea ice analysis (see below). IV.3.2 Validation results Since we do not know whether OSI SAF or the ice charts in general are most correct, we report only on the differences between the two. The figure below shows result from the validation from the 1 st quarter of The latest validation report is found at: EU Copernicus Marine Service Public Page 21/ 58

22 Figure 4. Comparison between OSI SAF ice concentration (= SEAICE_GLO_SEAICE_L4_NRT_- OBSERVATIONS_011_001 / edge dataset) and DMI ice charts (= SEAICE_ARC_SEAICE_L4_NRT_OBSERVATIONS_011_003) Validation of the overview charts and the regional charts give the same results. EU Copernicus Marine Service Public Page 22/ 58

23 IV.4 Regional Sea Ice Baltic Product Product description Production unit, PU Dissemination unit DU SEAICE_BAL_SEAICE_L4_NRT_ OBSERVATIONS_011_004 SEAICE_BAL_SEAICE_L4_NRT_ OBSERVATIONS_011_011 Baltic sea ice L4 SIW-FMI-HELSINKI-FI SIW-METNO-OSLO- NO Baltic sea ice SAR SIW-FMI-HELSINKI-FI SIW-METNO-OSLO- NO The regional Sea Ice products from Baltic are produced and delivered by FMI. These are ice drift, ice concentration and ice thickness. IV.4.1 Validation Procedure FMI validates ice thickness data, ice concentration data, and ice drift data. Ice thickness is validated with ice thickness measurements made on board Finnish icebreakers. Ice concentration validation is made against the ASI algorithm (by the University of Bremen). Ice drift data is validated against both drifter buoy data and drift measurements based on SAR images. The validation is performed twice for each season: the early winter validation in February (Dec-Jan data), and the late winter validation in May (Feb-May data), this six-month period covers a typical Baltic ice season rather well. The ice drift is validated only once for the whole season. The reason for this is that the ice drift estimation is started later than the other products, because the algorithm only works for thick enough ice to at least partially preserve its shape during drifting. Thus we have to wait until there exist ice patches which do not exhibit a complete metamorphosis during the drifting, and can be tracked from a SAR image to another. IV.4.2 Validation results Ice concentration As the ice charts use 10% classes for the ice concentrations, the ice concentration subtractions between Finnish (FI) and Bremen (BR) ice concentration data were also classified in to 10% classes from [-100,-90[ to [90,100]. The figure below shows the ice concentration distribution of the subtractions at individual dates and the mean distribution. Positive values mean that FI product has provided higher concentrations than the BR product and vice versa. Therefore the fatter right tail indicates that the BR product gives lower concentration values as the FI-product. However, the majority of the subtractions are between -10% and 10%. Hence, the FI product and BR product results provide rather similar results. EU Copernicus Marine Service Public Page 23/ 58

24 Figure 5: Distributions of the ice concentration subtractions between the FMI product and the BR product. Positive values indicate that FI product is providing higher concentrations than the BR product and vice versa. At x-axis are the class centers of the 10% wide classes and y-axis values are in percentages of the number of data. Ice thickness The ice chart based ice thickness results were compared with ice thickness results measured on icebreakers. The figure below shows the location of the icebreaker measurements in the Gulf of Finland. EU Copernicus Marine Service Public Page 24/ 58

25 Figure 6: Locations of the used icebreaker measurements and measured ice thicknesses in centimeters. The blue line represents the coast of the Gulf of Finland. The figure below shows ice thickness values from ice charts compared to measurements from the icebreakers. EU Copernicus Marine Service Public Page 25/ 58

26 Figure 7: Ice thickness values from ice charts and ice breaker measurements in centimeters. Solid line is the fitted regression line by the least squares method The SAR-based ice thickness results were compared with ice thickness results measured on icebreakers. Several icebreaker measurements were excluded, because no matching SAR image was available. EU Copernicus Marine Service Public Page 26/ 58

27 Figure 8: SAR based ice thicknesses and ice breaker measurements in centimeters. Solid line is the fitted regression line by the least squares method. Ice drift The ice drift monitoring period was from Jan 1st to Mar 19th All except one the buoys were in the Gulf of Bothnia, for details of the buoy trajectories, see figure below. EU Copernicus Marine Service Public Page 27/ 58

28 Figure 9: Buoy trajectories (buoys indicated by different colours) during the winter study period in One buoy not shown here was drifting in the western of Gulf of Finland. During the monitoring period totally 117 ice drift estimates were used in the comparison, i.e. the two buoy locations was within the overlapping area of the two adjacent SAR images used in the drift estimation for the acquisition times of the SAR images. The ice drift estimates in comparison were divided into two categories: 68 of these were classified to short drift category (buoy motion less than 500m during the time interval between two adjacent SAR images over the same area) and 49 measurements to long drift. The quality given by the SAR algorithm for the short drift data varied from 40% to 80% in both short drift and long drift categories. For the short drift data category only the motion magnitude was evaluated but for the long drift data both magnitude and direction were estimated. The direction couldn t be evaluated in the short drift category because in short drift estimates the SAR registration errors can cause large relative errors and so defining the direction can become ambiguous due to the quantification of the direction. For the long drift data both the magnitude and direction were combined. For the short drift data the L1 errors were 2207m (quality=40%, 63 samples), 5391m (quality=60%, 2 samples), and 237m (quality=80%, 3 samples). For the long drift the L1 errors were 4692m (quality=40%, 46 samples), and 225 (quality=80%, 3 samples). The time periods vary according to the acquisition times of the overlapping SAR image pairs, the maximum time is two and half days (60 hours). EU Copernicus Marine Service Public Page 28/ 58

29 The corresponding errors for the direction were degrees and degrees, indicating that the direction is still estimated moderately in these difficult conditions. The estimation accuracy of the motion magnitudes is not very good, except for the high quality value (80%) data. This may be due to that the buoys were located in the marginal ice zone. Figure 10: Magnitude comparison for the long motion, the values are in meters. The quality class 40% is coloured red, quality class 60% green (no points for the long motion data), and quality class 80% blue (3points). EU Copernicus Marine Service Public Page 29/ 58

30 Figure 11: Direction comparison for the long motion, the values are in degrees (directions 0 and 360 are the same direction). The quality class 40% is colored red (46 points), quality class 60% green (no points for the long motion data), and quality class 80% blue (3 points). EU Copernicus Marine Service Public Page 30/ 58

31 Figure 12: Magnitude comparison for the short motion, the values are in meters. The quality class 40% is colored red (63 points), quality class 60% green (2 points), and quality class 80% blue (3 points). ). Most of the points are concentrated near the origin, but there also seem to be some erroneous points along the y-axis (estimated SAR motion), these may be near the boundary of drift ice and static ice. The latest validation report is found at: IV.5 Arctic and Antarctic ice drift Product Product description Production unit, PU Dissemination unit DU SEAICE_GLO_SEAICE_L4_NRT_ OBSERVATIONS_011_006 Global high-resolution SAR sea ice drift SIW-DTUSPACE- COPENHAGEN-DK SIW-METNO-OSLO- NO SAR derived drift datasets are produced at the DTU Space and delivered in near real time to the Dissemination Unit at MET Norway. The drift product has been enhanced in V4 by including information from the automated monthly product validation in the product files. The quality information numbers describe the quality of the product by comparing one month of sea ice drift values to high quality GPS positions of drift buoys. The result of the comparison is the average value and the standard deviation of the differences between the product (sea ice drift) and the control (buoy GPS positions) located close in space and time. IV.5.1 Validation procedure The products are validated against reliable time series from buoys/drifters when such data are available. As part of the validation the product has been compared to high quality positions from in-situ drifters (GPS equipped Ice Tethered Profilers ITPs). EU Copernicus Marine Service Public Page 31/ 58

32 The Ice-Tethered Profiler data were collected and made available by the Ice-Tethered Profiler Program based at the Woods Hole Oceanographic Institution ( The validation results are given as error estimates (bias/standard deviation, scatterplots) of the X and Y components of the drift numbers. IV.5.2 Validation results Below are the results of a comparison of the drift product with ITP positions in the months December 2010 until end of March of 2011: Table 4: Sea ice drift comparison against ITP positions. EU Copernicus Marine Service Public Page 32/ 58

33 Figure 13: SAR ice drift compared to in-situ observation itp37 EU Copernicus Marine Service Public Page 33/ 58

34 Typical numbers for the standard deviation should be around 300m during the winter and 500m during summer, and low tens of meters for the average values. Latest validation results are found at IV.6 Arctic Ice berg density Product Product description Production unit, PU Dissemination unit DU SEAICE_ARC_SEAICE_L4_NRT_ OBSERVATIONS_011_007 Arctic SAR iceberg concentration SIW-DMI- COPENHAGEN-DK SIW-METNO-OSLO- NO The iceberg number density product is based on satellite data from SAR (Sentinel-1 and Radarsat-2). The output is iceberg number density in open-water areas (i.e. areas where no sea-ice is present) given as number of icebergs sampled in grid cells each covering 10 x 10 km. Each file produced is a netcdf file including a grid that more or less covers the entire Greenland Waters; however the spatial coverage of each SAR scene is much smaller about 400 x 400 km and 500 x 500 km for Sentinel-1 and Radarsat-2, respectively. IV.6.1 Validation procedure: Due to different forms of noise affecting the SAR signal, some false icebergs may be detected in the SAR imagery. However, due to the lack of (and difficulties in getting) ground truth data, true validation of iceberg number density is a challenging if not impossible. Thus, to describe whether the results are likely to be within a realistic range we have used about 10 years of SAR data from the DMI archive to generate iceberg statistics for the Greenland Waters. Each product is then compared to these statistical data. For each valid 10x10 km grid cell over the Greenland Waters, percentiles (P84 and P97) of iceberg number density have been derived. The percentiles have been derived using a running seasonal analysis. Assuming a normal distribution P84 approximately corresponds to the mean + 1 std. dev. and P97 to the mean + 2 std. dev. Although iceberg number density is not normally distributed we have used these percentiles to divide our observations into three categories of uncertainty: 1. normal (< P84), 2. critical (]P84:P97]), 3. extreme (]P97:P100]). Note that uncertainty data will (of course) only be available in areas with valid statistics. Error statistics are calculated assuming that all observations with number of icebergs per grid-cell larger than the 97 percentile (about mean + 2xSTD, assuming a normal distribution) may be potential errors. Note that the new mosaicked dataset currently does not include statistical layers in the netcdf files. For this product the implementation of these layers is planned to be done when the Sentinel archives include multiple years, and when the new sea ice scheme has been run on existing data through reanalysis. Also note that when more and more data become available in the archives the derived statistics will become more robust and we might be able also to include yearly time-trends in validation statistics for future upgrades. IV.6.2 Validation results The figure below shows two examples of iceberg number densities including category of uncertainty (the data was inferred from Radarsat-SAR on January 22nd, 2011 and March 4th, respectively). The January 22nd data has virtually no potential errors, whereas the data from March 4th has quite a lot of potential errors. These errors are most likely due to extremely low backscatter to the east of the sea ice present along the East Greenland coast on March 4th. The low backscatter results in very low image dynamics, which increases the significance of even small variations in backscatter that may be due to winds other met-ocean phenomena or data-related speckle-noise. Such conditions complicate iceberg detection. Results show that when errors occur, then typically a few false iceberg-targets appear in the data, however in very rare cases severe noise phenomena may lead to a huge amount EU Copernicus Marine Service Public Page 34/ 58

35 of false targets appearing. Quantification of these results is given in the table (Estimated Accuracy Numbers) in section I.3. Figure 14: Examples of iceberg number density data inferred from satellite borne SAR including categories of uncertainty. Top: illustration of data with virtually no potential errors. Bottom: illustration of data with a larger amount of potential errors. The latest validation report is found at: IV.7 Global ice concentration time series Product Product description Production unit, PU Dissemination unit DU SEAICE_GLO_SEAICE_REP_OB SERVATIONS_011_009 Ice concentration time series SIW-METNO-OSLO-NO SIW-METNO-OSLO- NO EU Copernicus Marine Service Public Page 35/ 58

36 The OSI SAF reprocessed sea ice concentration products are available from to except for shorter gaps due to satellite malfunction or planned maintenance. The SMMR instrument was operated every second day. IV.7.1 Validation procedure The data has been validated against operational sea ice charts from National Ice Center (NIC). The ice charts, intended for aiding navigation are produced on a regular basis covering all seasons, both Southern and Northern hemispheres and the time series cover the entire reanalysis period except for the period 1995 to 2002 on the southern hemisphere where we have been unable to acquire digital ice charts. The OSI SAF ice concentration is compared with the SIGRID CT ice concentration code of the ice charts. Where the CT code defines an ice concentration interval the average of the interval bounds is used in the comparison. The ice charts are compared with OSI SAF EASE 25 km ice concentration product. The parameters shown in the validation plots are defined as follows. The ice chart analysis ice concentration will be referred as IAC and OSI SAF ice concentration as OSIC: Table 5: The table shows parameters validated IV.7.2 Validation results Comparison between National Ice Center (NIC) ice charts for Northern Hemisphere and the OSI SAF reanalysis shows a clear seasonal cycle with 80% to 90% of cases meeting the criteria during winter and only 20% to 60% during the peak of summer melt EU Copernicus Marine Service Public Page 36/ 58

37 Figure 15: Ice concentration match between NIC and OSI SAF The bias in ice concentration between the Northern Hemisphere National Ice Center ice charts and OSI SAF reanalysis ice concentration is shown below. The OSI SAF reanalysis ice concentration is higher than the ice chart over open water. This is due to the fact that the radiometer ice concentration is affected by atmospheric noise which increases the ice concentration above zero. The ice chart has a nominal value of zero over open water. Figure 16: Comparison between the Northern Hemisphere National Ice Center ice charts and OSI SAF reanalysis ice concentration. The total bias between OSI SAF the reanalysis product and SEAICE_GLO_SEAICE_REP_OBSERVATIONS_011_009 the ice chart is shown with the blue curve, the negative bias in ice covered regions with the red curve and the positive bias in water areas with the yellow curve. The standard deviation of the difference between the OSI SAF reanalysis product and the National Ice Center ice charts. There is a clear seasonal cycle with higher standard deviations during summer than during winter. The standard deviation over open water seems to decrease during the reanalysis period. EU Copernicus Marine Service Public Page 37/ 58

38 Figure 17: Comparison between Northern Hemisphere National Ice Center ice charts and the OSI SAF reanalysis ice concentration. The figure shows the standard deviation the difference between the OSI SAF reanalysis product and the ice charts. The blue curve shows the standard deviation of the difference for both ice and open water regions. The red and the yellow curve show standard deviation of the difference for ice and water regions respectively. Validation results are found on A validation report is found at: IV.8 Arctic ice drift time series Product Product description Production unit, PU Dissemination unit DU SEAICE_ARC_SEAICE_REP_OB SERVATIONS_011_010 Arctic sea ice drift time series SIW-IFREMER-BREST- FR SIW-IFREMER- BREST-FR The product is based on a scatterometers and radiometer sea ice drift data since The synergy of both sensors ice drift product enables to have an extented estimation period, a higher reliability of the product than individual ones, and more vectors estimated. IV.8.1 Validation procedure: The sea ice drift have been compared with buoys from IABP (US). U and V components have been analysed through five winters (magnitude and direction). IV.8.2 Validation results The standard deviation of the difference to buoys is 7.0 km/29 for 3 day lags drift and 8.2 km/24 for 6 day lag drifts. The figure shows the angle difference between the product and buoys as a function of merged drift magnitudes: the angle difference sharply decreases (smaller than 45 ) for drift magnitudes higher than 40 km (about three pixels). This result in drift direction is comparable to previous validation results. The six-day lag is better for small drift magnitude estimation, and consequently, the angle data have a better resolution with a standard deviation. EU Copernicus Marine Service Public Page 38/ 58

39 The validation can be found at IV.9 Antarctic high-resolution ice edge Product Product description Production unit, PU Dissemination unit DU SEAICE_ANT_SEAICE_L4_NRT_ OBSERVATIONS_011_012 Antarctic high-resolution ice edge SIW-BAS- CAMBRIDGE-UK SIW-METNO-OSLO- NO The high-resolution regional sea ice edge product for the Antarctic is provided by BAS. It covers the Antarctic Peninsula and Weddell Se, Bellingshausen Sea, Amundsen Sea and eastern Ross Sea, provided every working day depending on availability of SAR image data. The temporal coverage is variable and focused on the austral summer season (October to March). The product is based on manual interpretation of satellite data from Radarsat-2 ScanSAR Wide and Sentinel-1 Extra Wide Mode imagery with a swath width of 500km, resampled to a spatial resolution of 100 meters. The nominal temporal span between interpreted swaths is 24 hours and the nominal product grid resolution is 1km. The final products are converted to and published in the NetCDF file format. IV.9.1 Validation procedure: Validation of the product is difficult due to lack of in situ observations and coincident high-resolution imagery for comparison. As an alternative the product will continuously compared with a manually interpreted sea ice edge from MODIS satellite imagery and the OSI SAF sea ice edge product. The products are compared using methodology established during an Eumetsat OSI SAF visiting scientist scheme (CDOP-SG07-VS01) comparing the CMEMS sea ice edge product with the OSI SAF sea ice edge product. To account for similar resolution of the MODIS and SAR images, the comparisons were performed on a 1 km grid. The statistical parameters derived from this analysis include: Average pixel classification agreement Average distance from CMEMS ice edge to MODIS ice edge EU Copernicus Marine Service Public Page 39/ 58

40 Average distance from MODIS ice edge to CMEMS ice edge IV.9.2 Validation results An example of the MODIS product and corresponding ice edge product are provided below: Figure 18: MODIS RGB colour composite image based on channel Results from the validation are shown in the table below. In general there is a good agreement as can be seen from the summary. Parameter Jan Feb Mar Jun Jul Sample number per month Average pixel classification agreement (%/100) Average distance from CMEMS ice edge to OSI SAF ice edge (pixels, 10 km pixel size) Table 6: The table gives a summary of the validation of the Antarctic high-resolution ice edge product. The latest validation report is found at: EU Copernicus Marine Service Public Page 40/ 58

41 V Appendix Appendix V.1 describes a study carried out during the MyOcean-2 project by the ice charting Production Units of the MyOcean OSI-TAC Workpackage. The same organisations are now delivering to the CMEMS OSI TAC service, so the results are fully applicable to the current CMEMS products of the same names. V.1 Ice Chart Uncertainty Estimates High resolution ice charts produced by the national Sea Ice Services of Denmark, Finland and Norway are in MyOcean made available to MFCs in a NetCDF format that are usable for validation of models and sea ice data assimilation. Uncertainty estimates are lacking in practically all fundamental ice chart data sets available. Therefore, a main task in MyOcean-2 has been to investigate the possibility for development of common / standardized validation methods for the high resolution sea ice analysis, with the aim of enabling the computation of consistent and comparable ice chart uncertainty estimates and error statistics, in terms of possible biases, variance and true resolution of the ice charts. Since there is not yet a better automatic tool for interpretation of sea ice in high resolution imagery, and due to the lack of field observations, ice charts are considered being the data sets that best describes the true state of the sea ice, and thus other sea ice data sets uses these as reference. The intercomparison of ice chart data sets and the analysis of differences is a useful way to evaluate the realism of the data sets and thus some estimates of ice chart uncertainty. This study seeks to do a systematic and comprehensive cross comparison of existing ice chart products from multiple MyOcean ice chart data sets over the same areas, as well as a comparison with the OSI SAF ice concentration (IC) data set (also available through MyOcean), to characterize important temporal and spatial differences in the data sets. First approach has been to compare the ice chart IC data, on full scale; all data, all common areas. Then, look at temporal and spatial differences; winter/summer data, individual years of data, local area comparisons. DMI has done comparison analysis of DMI Greenland Ice charts (SEAICE_ARC_SEAICE_L4_NRT_OBSERVATIONS_011_003) and MET Norway Svalbard Ice charts (SEAICE_ARC_SEAICE_L4_NRT_OBSERVATIONS_011_002), both covering Svalbard and Greenland waters. An additional comparison of the DMI and MET Norway Ice charts with OSI SAF IC data set (SEAICE_GLO_SEAICE_L4_NRT_OBSERVATIONS_011_001) was made for reference. This work has been done with input and in constructive discussion with MET Norway. Concurrently, FMI has done a similar comparison of FMI Ice charts (SEAICE_BAL_SEAICE_L4_NRT_OBSERVATIONS_011_004) and SHMI Ice charts both covering the Baltic Sea. The main results from these comparison analyses are presented in the sections below. These results were also presented and discussed at the 4th Ice Analyst Workshop (IAW-4) arranged by The International Ice charting Working Group (IICWG) in Helsinki, Finland 9th- 13th June In connection to these presentations an Ice Concentration Exercise was conducted at the workshop among the Ice Analyst and the results of that exercise are also presented in the section below. EU Copernicus Marine Service Public Page 41/ 58

42 V.1.1 Comparison of ice charts and OSI SAF ice concentration dataset in the Arctic DMI (Danish Meteorological Institute) and MET Norway (Norwegian Meteorological Institute) both publish ice charts covering Svalbard and Greenland waters. DMI produces an Overview Ice chart on Mondays and Thursday and Regional Ice charts covering different sub-areas on an irregular basis (SEAICE_ARC_SEAICE_L4_NRT_OBSERVATIONS_011_003). MET Norway produces a Svalbard Ice chart (SEAICE_ARC_SEAICE_L4_NRT_OBSERVATIONS_011_002), on every weekday. Through the MyOcean Online Data Catalogue, both the DMI and MET Norway ice chart IC data sets are available in NetCDF format in a 1 km grid (nominal resolution of the ice charts). The comparisons were made for the common area of the DMI and MET NORWAY ice charts for the whole season. Additionally, the same comparison analysis was made between the DMI and MET NORWAY ice chart data sets and the OSI SAF IC data set (SEAICE_GLO_SEAICE_L4_NRT_OBSERVATIONS_011_001) that covers the whole Arctic. This data set is also available through the MyOcean Online Data Catalogue in NetCDF format in a 10 km grid. Comparisons were also made on seasonal basis (summer, winter) and on separate years. Also, a local comparison in the Northeast Greenland area was made. Only the results of the comparison of the DMI ice charts, the MET NORWAY ice charts and the OSI SAF IC dataset for the full season are shown here. The parameter compared in this study was ice concentration. The data time series used in the comparison, are what was available in the catalogue up till November Results of the mentioned comparisons are shown in figures below. DMI Greenland Overview Ice Chart DMI Greenland Regional Ice Chart (example) MET Norway Svalbard Ice Chart OSISAF Ice Concentration dataset The approach has been, for each data set comparison, to find the data samples that are produced on the same day and to compare ice concentration values (and also ice thickness values for the Baltic ice charts) on grid cell level. The data sets have different characteristics that are described in the corresponding PUMs. There are two sets of results for each data set comparison, where each of the datasets serves as the basis for the comparison. Results are plotted in a blurred scatter plot that visualizes the frequency of the data values; white numbers are the occurrence of data values in (rounded) percent (of the total number of corresponding ice concentration values). Zero indicates an occurrence of less than half a percent. A table next to the graph show statistics mean, median and standard deviation - for each unique ice concentration in the dataset on the value axis. The large number of corresponding open water (0% ice concentration) grid cells in the data sets has been neglected in the comparisons, so that only corresponding grid cells where there is ice in one of the data sets, is taking into account. If the many corresponding open water grid cells were included in the analysis the results for 0% ice concentration would undoubtedly have a mean of 0% and an insignificant standard deviation). DMI overview ice DMI overview ice MET NORWAY DMI regional ice chart Comparison chart chart Svalbard ice chart and MET NORWAY analyses And MET NORWAY and and Svalbard ice chart Svalbard ice chart OSISAF IC dataset OSISAF IC dataset Data time series EU Copernicus Marine Service Public Page 42/ 58

43 Analysis with overlap IC grid cells 9.5 M 66.4 M 2.5 M 16.4 M Table 3: General information on dataset intersection. (M = million). DMI MET NORWAY IC Mean Std.dev. Median MET NOR DMI WAY IC Mean Std.dev. Median Figures above: Scatter density plots of [top] DMI regional ice charts IC and corresponding MET NORWAY ice charts IC [bottom] MET NORWAY ice charts IC and corresponding DMI regional ice charts IC, Black dots are mean values. Vertical lines are standard deviations. White numbers are occurrence in percent, rounded. Solid black line is a reference line (x=y). Tables above: Statistics based on the data also shown in the figures to the left. EU Copernicus Marine Service Public Page 43/ 58

44 DMI MET Norway IC Mean Std.dev. Median MET DMI IC Mean Std.dev. Median Figures above: Scatter density plots of [top] DMI overview ice charts IC and corresponding MET NORWAY ice charts IC and [bottom] MET NORWAY ice charts IC and corresponding DMI overview ice charts IC, Black dots are mean values. Vertical lines are standard deviations. White numbers are occurrence in percent, rounded. Solid black line is a reference line (x=y). Tables above: Statistics based on the data also shown in the figures to the left. EU Copernicus Marine Service Public Page 44/ 58

45 An additional analysis has been made, to investigate the difference in ice concentrations in an ice chart from each ice service, compared to the temporal evolution in ice concentrations of two subsequent ice charts from the same ice service. Here, the difference between ice concentrations in a MET NORWAY ice chart and a DMI overview ice chart from the same date is compared to the difference in the corresponding ice concentrations values in two subsequent Greenland overview ice charts. Figure: Scatter density plot of the IC difference between MET NORWAY ice charts and corresponding DMI overview ice charts from the same day (t1) and the IC difference between two subsequent DMI overview ice charts (t1 to t2), for the period Red dots are mean values. Black numbers are occurrence (of grid cell pairs) in percent times ten, rounded. EU Copernicus Marine Service Public Page 45/ 58

46 DMI OSI SAF IC Mean Std.dev. Median OSISAF DMI IC Mean Std.dev. Median Figures above: Scatter density plots of [top] DMI overview ice charts IC and corresponding OSI SAF IC and [bottom] OSI SAF IC and corresponding DMI overview ice charts IC, Black dots are mean values. Vertical lines are standard deviations. White numbers are occurrence in percent, rounded. Solid black line is a reference line (x=y). Tables above: Statistics based on the data also shown in the figures to the left. EU Copernicus Marine Service Public Page 46/ 58

47 MET OSI SAF IC Mean Std.dev. Median OSISAF MET IC Mean Std.dev. Median Figures above: Scatter density plots of [top] MET Norway ice charts IC and corresponding OSI SAF IC and [bottom] OSI SAF IC and corresponding MET Norway ice charts IC, Black dots are mean values. Vertical lines are standard deviations. White numbers are occurrence in percent, rounded. Solid black line is a reference line (x=y). Tables above: Statistics based on the data also shown in the figures to the left. Conclusion on comparison of ice charts in the Arctic EU Copernicus Marine Service Public Page 47/ 58