Global Sea Ice Edge and Type Validation...

28
Ocean & Sea Ice SAF Global Sea Ice Edge and Type Validation Report GBL SIE OSI-402-b and GBL SIT OSI-403-b Version 1.1 — April 2015 Signe Aaboe, Lars-Anders Breivik, Steinar Eastwood and Atle Sørensen Norwegian Meteorological Institute

Transcript of Global Sea Ice Edge and Type Validation...

Page 1: Global Sea Ice Edge and Type Validation Reportosisaf.met.no/docs/osisaf_cdop2_ss2_valrep_ice-edge-type_v1p1.pdf · diance (DLI). The sea ice products include sea ice concentration,

Ocean & Sea Ice SAF

Global Sea Ice Edge and TypeValidation Report

GBL SIE OSI-402-b and GBL SIT OSI-403-b

Version 1.1 — April 2015

Signe Aaboe, Lars-Anders Breivik, Steinar Eastwoodand Atle Sørensen

Norwegian Meteorological Institute

Page 2: Global Sea Ice Edge and Type Validation Reportosisaf.met.no/docs/osisaf_cdop2_ss2_valrep_ice-edge-type_v1p1.pdf · diance (DLI). The sea ice products include sea ice concentration,
Page 3: Global Sea Ice Edge and Type Validation Reportosisaf.met.no/docs/osisaf_cdop2_ss2_valrep_ice-edge-type_v1p1.pdf · diance (DLI). The sea ice products include sea ice concentration,

Documentation Change Record:

Version Date Author Description1.0 30.01.2015 SAA Initial version1.1 10.04.2015 SAA Amended according to comments

from the ice review

Page 4: Global Sea Ice Edge and Type Validation Reportosisaf.met.no/docs/osisaf_cdop2_ss2_valrep_ice-edge-type_v1p1.pdf · diance (DLI). The sea ice products include sea ice concentration,

SAF/OSI/CDOP2/MET-Norway/SCI/RP/224

Table of contents

Table of contents

1 Introduction 11.1 The EUMETSAT Ocean and Sea Ice SAF . . . . . . . . . . . . . . . . . . . . 11.2 Scope . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11.3 Overview . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11.4 Glossary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3

2 Validation data set 42.1 Ice charts from Northern Hemisphere . . . . . . . . . . . . . . . . . . . . . . 42.2 Ice charts from Southern Hemisphere . . . . . . . . . . . . . . . . . . . . . . 7

3 Validation methodology 83.1 Sea ice edge . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8

3.1.1 The match between ice products by pixels . . . . . . . . . . . . . . . . 93.1.2 The mean distance between product ice edges . . . . . . . . . . . . . 93.1.3 Target accuracy for sea ice edge . . . . . . . . . . . . . . . . . . . . . 10

3.2 Sea ice type . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 103.2.1 Monitoring of the ice type variability . . . . . . . . . . . . . . . . . . . 103.2.2 Target accuracy for sea ice type . . . . . . . . . . . . . . . . . . . . . 11

3.3 Comparison with operational static algorithm . . . . . . . . . . . . . . . . . . 11

4 Validation results and discussion 124.1 Validation of sea ice edge . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12

4.1.1 Northern Hemisphere . . . . . . . . . . . . . . . . . . . . . . . . . . . 124.1.2 Southern Hemisphere . . . . . . . . . . . . . . . . . . . . . . . . . . . 16

4.2 Validation of sea ice type . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 204.2.1 Northern Hemisphere . . . . . . . . . . . . . . . . . . . . . . . . . . . 20

5 Conclusion 22

References 24

EUMETSAT OSI SAF Version 1.1 — April 2015

Page 5: Global Sea Ice Edge and Type Validation Reportosisaf.met.no/docs/osisaf_cdop2_ss2_valrep_ice-edge-type_v1p1.pdf · diance (DLI). The sea ice products include sea ice concentration,

SAF/OSI/CDOP2/MET-Norway/SCI/RP/224 1

1. Introduction

1.1 The EUMETSAT Ocean and Sea Ice SAF

For complementing its Central Facilities capability in Darmstadt and taking more benefit fromspecialized expertise in Member States, EUMETSAT created Satellite Application Facilities(SAFs), based on co-operation between several institutes and hosted by a National Meteo-rological Service. More on SAFs can be read from www.eumetsat.int.

The Ocean and Sea Ice Satellite Application Facility (OSI SAF) is producing on an opera-tional basis a range of air-sea interface products, namely: wind, sea ice characteristics, SeaSurface Temperatures (SST), Surface Solar Irradiance (SSI) and Downward Longwave Irra-diance (DLI). The sea ice products include sea ice concentration, the sea ice emissivity, seaice edge, sea ice type and sea ice drift and sea ice surface temperature (from mid 2014).

The OSI SAF consortium is hosted by Météo-France. Sea ice products are produced at theOSI SAF High Latitude processing facility (HL centre), operated jointly by the NorwegianMeteorological Institute (MET-Norway) and Danish Meteorological Institute (DMI). Resultingsea ice fields are available daily within 6 hours after the last satellite data acquisition. Thismeans within 06 UTC each day. The sea ice products are delivered with global coverage ontwo files, on for the Northern and on for the Southern Hemisphere. In addition a separateproduct for the Atlantic part of the northern Hemisphere is produced.

Note: The ownership and copyrights of the data set belong to EUMETSAT. The data is dis-tributed freely, but EUMETSAT must be acknowledged when using the data. EUMETSAT’scopyright credit must be shown by displaying the words "copyright (year) EUMETSAT" oneach of the products used. User feedback to the OSI SAF project team is highly valued. Thecomments we get from our users is important argumentation when defining developmentactivities and updates. We welcome anyone to use the data and provide feedback.

1.2 Scope

The purpose of this report is to present validation results for the EUMETSAT OSI SAF Globalproducts of sea ice edge (OSI-402-b) and type products (OSI-403-b).

1.3 Overview

The global sea ice edge and sea ice type products are classification products here givenwith their associated quality flags:

• Sea ice edge (OSI-402-b) - distinguish between open water, open sea ice and closedsea ice

EUMETSAT OSI SAF Version 1.1 — April 2015

Page 6: Global Sea Ice Edge and Type Validation Reportosisaf.met.no/docs/osisaf_cdop2_ss2_valrep_ice-edge-type_v1p1.pdf · diance (DLI). The sea ice products include sea ice concentration,

SAF/OSI/CDOP2/MET-Norway/SCI/RP/224 2

• Sea ice type (OSI-403-b) - distinguish between first-year sea ice and multi-year seaice.

Both products are multi-sensor products derived from passive microwave (SSMIS) and ac-tive microwave scatterometer (ASCAT) data combined in a Bayesian approach. They arecomputed for both hemispheres on the standard OSI SAF grid with 10 km spatial resolution.

Note: Due to uncertainties regarding melting and wet ice during summer months, theOSI SAF sea ice type has meaningful data in the period from mid-October to mid-May,whereas the period from mid-May to mid-October is classified as ambiguous in this prod-uct. Also note, that at present the OSI SAF sea ice type product delivered for SouthernHemisphere classify all sea ice as ambiguous.

Todays operational algorithm is based on static background statistics. The present upgrade,which is validated in this report, is due to the inclusion of a dynamical algorithm in orderto improve transitions between seasons as well as sensors. So when we in this reportare referring to the dynamical algorithm or product we mean the new algorithm upgrade.Similarly, when referring to the static algorithm we mean the todays operational version ofthe algorithm. More detailed information on the algorithm - dynamical and static, the inputsatellite data and technical issues regarding the products, are given in the correspondingProduct User’s Manual [PUM, Aaboe et al., 2015b] and the Algorithm Theoretical BasisDocument [ATBD, Aaboe et al., 2015a].

See http://osisaf.met.no for real time examples of the products and updated information.General information about the OSI SAF is given at http://www.osi-saf.org.

Section 2 presents the data set used for validation of the sea ice edge and Section 3 de-scribes the methodology used for validating the products. Results of the validation of theproducts and discussions are presented in Section 4 and Section 5 concludes the valida-tion.

EUMETSAT OSI SAF Version 1.1 — April 2015

Page 7: Global Sea Ice Edge and Type Validation Reportosisaf.met.no/docs/osisaf_cdop2_ss2_valrep_ice-edge-type_v1p1.pdf · diance (DLI). The sea ice products include sea ice concentration,

SAF/OSI/CDOP2/MET-Norway/SCI/RP/224 3

1.4 Glossary

AARI Arctic and Antarctic Russian InstituteAMSR-2 Advanced Microwave Scanning Radiometer - 2ASCAT Advanced SCATerometerATBD Algorithm Theoretical Basis DocumentCDOP Continuous Development and Operations PhaseDLI Downward Longwave IrradianceDMI Danish Meteorological InstituteDMSP Defense Meteorological Satellite ProgramESA European Space AgencyEUMETSAT European Organization for the Exploitation of Meteorological SatellitesHL High LatitudeMET-Norway Norwegian Meteorological InstituteMODIS Moderate-Resolution Imaging SpectroradiometerNH Northern HemisphereNIC U.S. National Ice CenterNOAA National Oceanic and Atmospheric AdministrationOSI SAF Ocean and Sea Ice SAFPDF Probability Density FunctionPMW Passive Micro WavePUM Product User’s ManualSAF Satellite Application FacilitySAR Synthetic Aperture RadarSH Southern HemisphereSSI Surface Solar IrradianceSSM/I Special Sensor Microwave/ImagerSSMIS Special Sensor Microwave Imager/SounderSST Sea Surface Temperaturestd standard deviationUTC Coordinated Universal TimeWMO World Meteorological Organization

EUMETSAT OSI SAF Version 1.1 — April 2015

Page 8: Global Sea Ice Edge and Type Validation Reportosisaf.met.no/docs/osisaf_cdop2_ss2_valrep_ice-edge-type_v1p1.pdf · diance (DLI). The sea ice products include sea ice concentration,

SAF/OSI/CDOP2/MET-Norway/SCI/RP/224 4

2. Validation data set

The OSI SAF sea ice edge product is continuously validated against ice charts originatingfrom the following operational ice chart divistions:

• DMI (http://www.dmi.dk/en/groenland/hav/ice-charts/)

• MET-Norway (http://polarview.met.no/)

• U.S. National Ice Service (NIC) (http://www.natice.noaa.gov/).

The ice charts will be used as “ground truth” against which the OSI SAF ice edge productwill be compared. As the ice charts are manual interpretation of satellite data (mainly SAR),they are not proper in situ data. The ice categories used are representative for an area,as the ice charter draws polygons. Even so, these charts are the best available source forvalidation.

2.1 Ice charts from Northern Hemisphere

At the ice charting division at DMI is produced ice charts covering Greenland Waters. Thevalidation against these charts is carried out at DMI on a weekly basis.

On a daily (weekdays) basis the Norwegian Ice Service at MET-Norway produces ice chartswhich cover the area from the Fram Strait to the Barents Sea with main emphasis on theSvalbard area. An example from the 3rd of November 2014 is shown in Figure 1. Forcomparison, the OSI SAF sea ice edge product for the 3rd November 2014 is shown inFigure 2.

The ice charts are to a large extent based on a subjective interpretation of high resolutionsatellite data, predominantly Sentinel-1 Extra-Wide and RADARSAT-2 ScanSAR Wide butalso some COSMO SkyMed, MODIS optical images, and AMSR-2 sea ice concentrationfrom University of Bremen. The detailed interpretation of satellite images and a subsequentmapping procedure are carried out by skilled (experienced and trained) ice analysts.

In areas where high resolution data are not available for subjective analysis the ice analystmight use SSMIS data and OSI SAF products. Therefore, for each ice chart, the ice analystselect target areas where detailed SAR data are available for their subjective interpretation.Only these areas are qualified as being an independent source from the OSI SAF ice prod-ucts and can be used in the validation. In the ice chart example in Figure 1, the black boxesrepresent such target areas which are used in the validation.

EUMETSAT OSI SAF Version 1.1 — April 2015

Page 9: Global Sea Ice Edge and Type Validation Reportosisaf.met.no/docs/osisaf_cdop2_ss2_valrep_ice-edge-type_v1p1.pdf · diance (DLI). The sea ice products include sea ice concentration,

SAF/OSI/CDOP2/MET-Norway/SCI/RP/224 5

Figure 1: Svalbard region ice chart dated 3rd November 2014 from the Norwegian Ice Ser-vice (http://polarview.met.no/). The colors, labeled in the lower left box, indicate the iceconcentration where red indicates highest concentration and blue the lowest concentration.Red lines represents isolines of the sea surface temperature. The regions marked with blackboxes indicate where detailed SAR (in this case Sentinel-1 and RADARSAT-2) data existedthat day.

EUMETSAT OSI SAF Version 1.1 — April 2015

Page 10: Global Sea Ice Edge and Type Validation Reportosisaf.met.no/docs/osisaf_cdop2_ss2_valrep_ice-edge-type_v1p1.pdf · diance (DLI). The sea ice products include sea ice concentration,

SAF/OSI/CDOP2/MET-Norway/SCI/RP/224 6

Figure 2: OSI SAF sea ice edge product in the Northern Hemisphere from the 3rd of Novem-ber 2014.

EUMETSAT OSI SAF Version 1.1 — April 2015

Page 11: Global Sea Ice Edge and Type Validation Reportosisaf.met.no/docs/osisaf_cdop2_ss2_valrep_ice-edge-type_v1p1.pdf · diance (DLI). The sea ice products include sea ice concentration,

SAF/OSI/CDOP2/MET-Norway/SCI/RP/224 7

2.2 Ice charts from Southern Hemisphere

The Southern Hemisphere OSI SAF sea ice edge product is validated against the bi-weeklyice charts produced at the U.S. National Ice Center.

In addition, currently MET-Norway provides a weekly Weddell Sea ice chart during the South-ern Hemisphere summer period - mainly between October and April, see an example inFigure 3. The Weddell Sea ice chart is a new validation source for the OSI SAF sea iceproducts and validation against these ice charts are included in this validation report, butare at present not included in the OSI SAF half-yearly operations report.

Figure 3: Weddell Sea ice chart produced on the 16th of March 2015 at the Norwegian IceService (http://polarview.met.no/). The colors, labeled in the lower left box, indicate the iceconcentration where red indicates highest concentration and blue the lowest concentration.Red lines represents isolines of the sea surface temperature. The regions marked with blackboxes indicate where detailed SAR (in this case Sentinel-1 and RADARSAT-2) data existedthat day.

A collaborative Antarctic sea ice product is being developed between the U.S. National IceCenter, Arctic and Antarctic Russian Institute (AARI), and Norwegian Ice Service (MET-Norway) which will provide a weekly comprehensive Antarctic chart throughout the year.The charts are expected to be fully developed by the end of 2015. This new collaborativeproduct will be continuous and will also include more high resolution products due to eachinstitutes ability to share efforts [P. Wagner, Norwegian Ice Service, pers. comm., April 14th2015].

EUMETSAT OSI SAF Version 1.1 — April 2015

Page 12: Global Sea Ice Edge and Type Validation Reportosisaf.met.no/docs/osisaf_cdop2_ss2_valrep_ice-edge-type_v1p1.pdf · diance (DLI). The sea ice products include sea ice concentration,

SAF/OSI/CDOP2/MET-Norway/SCI/RP/224 8

3. Validation methodology

The validation is twofolds: the continuous validation of the operational product, and thevalidation against historical data which is presented in this report. The method is the samefor both cases.

3.1 Sea ice edge

The comparison of the OSI SAF ice edge product against the ice charts is evaluated in twoways:

• The pixels within the validation area are compared, pixel by pixel, and the occurrencesof agreements/disagreements between the ice chart and the OSI SAF product arecounted.

• The mean distance between the two ice edges is calculated (in km)

The pixel-by-pixel comparison gives an overall validation of the product, while the distancebetween the two ice edges gives details of the most interesting area around the ice edge.

The real ice edge partly depends on the scale one is observing on and the ice charts havemuch higher resolution than the OSI SAF product. Therefore, before the comparison, theice chart is averaged onto the OSI SAF grid, which is a polar-stereographic grid with 10 kmspatial resolution.

The ice charts give ice concentration from 0% (no ice) to 100% (fully ice-covered sea),whereas the OSI SAF ice edge product has the classes open water, open ice, and closedice. In order to compare the two products a threshold must be drawn for the ice charts, de-termining what to classify as “ice” and what to call “ice free”. Here, ice concentrations higherthan or equal to 35% are considered to classify as “ice”, and the lower ice concentrationsare translated to “no ice”. The threshold of 35% corresponds to the boarder between theWMO sea ice nomenclature “very open drift ice” with ice concentration within the range 1

10to 3

10 and the “open drift ice” with concentration range 410 to 6

10 (WMO publication No. 259,Suppl. No. 4).

OSI SAF classes Ice chart concentrationice open ice, closed ice ≥ 35%no ice open water < 35%

EUMETSAT OSI SAF Version 1.1 — April 2015

Page 13: Global Sea Ice Edge and Type Validation Reportosisaf.met.no/docs/osisaf_cdop2_ss2_valrep_ice-edge-type_v1p1.pdf · diance (DLI). The sea ice products include sea ice concentration,

SAF/OSI/CDOP2/MET-Norway/SCI/RP/224 9

3.1.1 The match between ice products by pixels

Within the validation area a pixel-by-pixel comparison is carried out between the two griddedice products, and the occurrences of each of the four possibilities are counted:

1. no ice in OSI SAF and no ice in the ice chart

2. no ice in OSI SAF but ice in the ice chart

3. ice in OSI SAF but no ice in the ice chart

4. ice in OSI SAF and ice in the ice chart

In case 1 (no ice, no ice) and 4 (ice, ice) the OSI SAF product agrees with the ice chart. Theterm match is defined as the probability of “true” detection of ice or no ice:

match =N1 +N4

N· 100% (3.1)

Here, N1 and N4 represent the number of pixels with occurrence 1 and 4, respectively. N isthe total number of pixels used in the comparison, that is N = N1+N2+N3+N4. In order tosee if the OSI SAF sea ice cover has a bias toward either underestimating or overestimatingthe sea ice cover in the ice charts the following two terms for the OSI SAF product are alsodefined:

underestimate =N2

N· 100% (3.2)

overestimate =N3

N· 100% (3.3)

Here, N2 and N3 represent the number of pixels with occurrence 2 and 3, respectively. Thethree terms match, underestimate, and overestimate add up to 100%.

3.1.2 The mean distance between product ice edges

The following is done for each pixel within the validation area:

When a pixel in the ice chart is found to contain ice (according to the thresholdset above), the surrounding eight pixels are checked. If any of these eight belongto the category no ice, the ice chart ice edge is located. The next step is thento find the distance to the nearest (ice, no_ice) neighboring pair of pixels in theOSI SAF product. The nearest pixels are checked first, and then expanding topixels further away until an OSI SAF ice edge is found and the distance (dist) iscomputed.

EUMETSAT OSI SAF Version 1.1 — April 2015

Page 14: Global Sea Ice Edge and Type Validation Reportosisaf.met.no/docs/osisaf_cdop2_ss2_valrep_ice-edge-type_v1p1.pdf · diance (DLI). The sea ice products include sea ice concentration,

SAF/OSI/CDOP2/MET-Norway/SCI/RP/224 10

The mean edge difference is defined as the average of the calculated distances for all theedge-pixel:

mean_edge_difference =∑

i dist(i)

Nedge(3.4)

Here, dist(i) is the edge distance calculation for a single edge-pixel (i), and Nedge is the totalnumber of edge-pixels for which a distance is calculated. The mean_edge_difference isgiven in km.

Note, that if the status flag for a pixel in the OSI SAF ice edge product is anything other thannominal, the pixel is skipped and does not contribute to the validation. Pixels near land ornear areas of missing data, or pixels lacking ice information for other reasons are thereby notallowed to reduce the quality of the validation. Conference the PUM [Aaboe et al., 2015b]for a list of possible status flags.

At present, the edge distance validation is only carried out for MET-Norway ice charts, thatis the Svalbard region on Northern Hemisphere, and the Weddell Sea region on SouthernHemisphere.

3.1.3 Target accuracy for sea ice edge

The target accuracy for the OSI SAF sea ice edge product is defined in the OSI SAF CDOP-2Service Specification Document [OSI SAF project team, 2014]. In this document it is statedthat the OSI SAF global sea ice edge has the following accuracy requirements:

Target accuracy: 20 km distance to ice edge (yearly average)

3.2 Sea ice type

3.2.1 Monitoring of the ice type variability

The multi-year ice is assumed not to have rapid changes in the horizontal coverage, andtherefore the OSI SAF ice type product quality assessment is done as a monitoring of thetemporal variation of the multi-year ice area coverage.

The monitoring is in two steps:

1. Monitoring of the daily differences of the area extent of the multi-year ice from its 11-days running mean. According to the assumption of no rapid changes, these dailydifferences should not be too large in periods with normal data input.

2. Calculation of the monthly standard deviation (std) in the daily difference from step 1.

In days with spatial gaps due to missing satellite input data the estimates of the multi-year icearea can be influenced if the gaps are within this region. Therefore days where missing dataarea exceeds 200.000 km2 are removed before the monthly statistics are carried out. Themonthly std (in step 2) of the multi-year ice area coverage is used to monitor the variability

EUMETSAT OSI SAF Version 1.1 — April 2015

Page 15: Global Sea Ice Edge and Type Validation Reportosisaf.met.no/docs/osisaf_cdop2_ss2_valrep_ice-edge-type_v1p1.pdf · diance (DLI). The sea ice products include sea ice concentration,

SAF/OSI/CDOP2/MET-Norway/SCI/RP/224 11

of the detrended product where the 11-days running mean has been removed from the dailyarea extent.

3.2.2 Target accuracy for sea ice type

The target accuracy for the OSI SAF sea ice type product is defined in the OSI SAF CDOP-2Service Specification Document [OSI SAF project team, 2014] . In this document it is statedthat the global sea ice type has the following accuracy requirements:

Target accuracy: 100.000 km2 monthly std in difference from running mean.

3.3 Comparison with operational static algorithm

The present upgrade of OSI-402-b and OSI-403-b from OSI-402-a and OSI-403-a is basedmainly on introducing dynamical background statistics into the algorithm in order to improvetransitions between seasons. Therefore, a comparison of the dynamical products againstthe operational product based on static background statistics is included. For more detailsabout the upgraded algorithm see the ATBD [Aaboe et al., 2015a].

EUMETSAT OSI SAF Version 1.1 — April 2015

Page 16: Global Sea Ice Edge and Type Validation Reportosisaf.met.no/docs/osisaf_cdop2_ss2_valrep_ice-edge-type_v1p1.pdf · diance (DLI). The sea ice products include sea ice concentration,

SAF/OSI/CDOP2/MET-Norway/SCI/RP/224 12

4. Validation results and discussion

For the present validation of the OSI SAF global sea ice products, both sea ice edge andsea ice type have been reprocessed for the years 2010-2014 with the upgraded dynamicalgorithm. Results from the previous static algorithm are included in the results for compari-son.

4.1 Validation of sea ice edge

4.1.1 Northern Hemisphere

Figure 4 shows the monthly time series of the mean_edge_differences (Equation 3.4) com-pared with MET-Norway ice charts from around Svalbard. The time series shows variabilityduring the period with values between 8 km and almost 40 km at the most. There is a gen-eral trend of worse performance of OSI SAF ice edge product during Arctic summer months– May to September. In average, the summer mean_edge_differences are about a factor 2higher than the winter values.

(a) Mean_edge_difference [km]

(b) Nedge

Figure 4: Monthly time series of the mean_edge_difference validation around the Svalbardregion (see Equation 3.4). (a) The mean distance [km] between the OSI SAF sea ice edgeand the ice edge from the MET-Norway ice charts. (b) Nedge is the total number of edge-pixels used in the estimate.

Table 1 summarizes the mean_edge_differences in Figure 4 as an annual mean for each ofthe five reprocessed years. In the second column is shown the values resulting from the up-

EUMETSAT OSI SAF Version 1.1 — April 2015

Page 17: Global Sea Ice Edge and Type Validation Reportosisaf.met.no/docs/osisaf_cdop2_ss2_valrep_ice-edge-type_v1p1.pdf · diance (DLI). The sea ice products include sea ice concentration,

SAF/OSI/CDOP2/MET-Norway/SCI/RP/224 13

dated dynamical algorithm, while in the right column is shown the corresponding values fromthe previous static algorithm. As can be seen, each year fulfills the accuracy requirementof 20 km distance to the ice chart ice edge (see section 3.1.3). Despite the year of 2013,the dynamical algorithm gives rise to slightly higher mean_edge_differences than the staticalgorithm. By comparing with Figure 4 this difference is to a large extent caused during thesummer months.

Year Mean_edge_difference [km] Mean_edge_difference [km] Nedge

dyn stat2010 19.29 17.84 383832011 19.48 18.49 422462012 16.38 15.39 550322013 18.28 18.31 462482014 17.11 16.19 38552

Table 1: The annual mean distance between the OSI SAF ice edge and the MET-Norwayice chart edge (Northern Hemisphere). Second column shows the values resulting fromthe dynamical algorithm, while the third column shows the corresponding values from theprevious static algorithm. Nedge is the total number of edge-pixels used in the estimate.

Figure 5 shows the monthly time series of the pixel-by-pixel comparison [%] between OSI SAFsea ice edge and MET-Norway ice charts around Svalbard (notice, the different scale on they -axis). Figure 5a shows the match when the OSI SAF product agrees with the ice chart(see Equation 3.1). In general there is a better match-performance during autumn, winter,and spring, whereas the match has a drop in summer months. Figures 5b and 5c show theunderestimate and the overestimate, respectively, when the OSI SAF product disagrees withthe ice chart (see Equations 3.2 and 3.3). Of the two disagreement-terms, there seems to bea slight bias toward an underestimation of the sea ice edge. The underestimate has summerpeaks with higher values. This means the OSI SAF algorithm seems to underestimate theice edge during summer time. No similar seasonal trend is seen for the overestimate. In Fig-ures 5a-5c the validation of the “static” sea ice product is included for comparison (dashedlines). The magnitude of match are comparable for the two algorithms.

Figure 6 shows the weekly time series of the pixel-by-pixel comparison [%] between OSI SAFsea ice edge and the DMI ice charts around Greenland. In Figure 6a is the validation forthe reprocessed ice edge product based on dynamical PDF’s. Figure 6b is taken from theHalf-Yearly Operations Report (2nd half 2014) [OSI SAF project team, 2015, Fig.51] and isbased on static PDF’s in the previous algorithm. In generel, the validation around Green-land shows the same trends as for the Svalbard region: The OSI SAF sea ice edge producthas best performance during autumn, winter and spring; the disagreements are biased to-ward underestimation of the ice edge - with peaks in summer time; and the overestimationshows no clear seasonality during the last 10 years. Comparing Figures 6a and 6b showscomparable values for the static and the dynamical product.

EUMETSAT OSI SAF Version 1.1 — April 2015

Page 18: Global Sea Ice Edge and Type Validation Reportosisaf.met.no/docs/osisaf_cdop2_ss2_valrep_ice-edge-type_v1p1.pdf · diance (DLI). The sea ice products include sea ice concentration,

SAF/OSI/CDOP2/MET-Norway/SCI/RP/224 14

(a) Match [%] (see Equation 3.1)

(b) Underestimate [%] (see Equation 3.2)

(c) Overestimate [%] (see Equation 3.3)

(d) N

Figure 5: Monthly time series of the pixel-by-pixel validation of the OSI SAF sea ice edgeproduct against MET-Norway ice charts around the Svalbard region. The validation is givenin terms of (a) match, (b) underestimate and (c) overestimate which are defined in Sec-tion 3.1. (d) N is the total number of pixels used in the comparison.

EUMETSAT OSI SAF Version 1.1 — April 2015

Page 19: Global Sea Ice Edge and Type Validation Reportosisaf.met.no/docs/osisaf_cdop2_ss2_valrep_ice-edge-type_v1p1.pdf · diance (DLI). The sea ice products include sea ice concentration,

SAF/OSI/CDOP2/MET-Norway/SCI/RP/224 15

(a) Validation based on dynamical PDF’s for 2010-2014.

(b) Validation based on static PDF’s back to 2002 [OSI SAF project team, 2015, Fig.51].

Figure 6: Weekly time series of the pixel-by-pixel validation of the OSI SAF sea ice edgeproduct against DMI ice charts around Greenland. The validation is given in the termsmatch (light gray), underestimate (dark gray) and overestimate (black), which are defined inSection 3.1.

EUMETSAT OSI SAF Version 1.1 — April 2015

Page 20: Global Sea Ice Edge and Type Validation Reportosisaf.met.no/docs/osisaf_cdop2_ss2_valrep_ice-edge-type_v1p1.pdf · diance (DLI). The sea ice products include sea ice concentration,

SAF/OSI/CDOP2/MET-Norway/SCI/RP/224 16

4.1.2 Southern Hemisphere

Figure 7 shows the monthly time series of the mean_edge_differences (Equation 3.4) com-paring with MET-Norway ice charts in the Weddell Sea region for the Antarctic summers in2010-2014. The summer period normally covers October to April, with a few exceptions.The time series shows variability during the period with values from around 10 km to around90 km. Despite the variability there is a trend of peak values during mid-summer – Decem-ber to February. And again, as for the Northern Hemisphere, the dynamical product oftenperform slightly worse in summer than the static product.

(a) Mean_edge_difference [km].

(b) Nedge.

Figure 7: Monthly time series of the mean_edge_difference validation for the Weddel Searegion (see Equation 3.4). (a) The mean distance [km] between the OSI SAF sea ice edgeand the ice edge from the MET-Norway ice charts. (b) Nedge is the total number of edge-pixels used in the estimate.

Table 2 summarizes the mean_edge_differences in Figure 7 as an annual mean using onlythe available summer months for each of the five reprocessed years. In the second col-umn is shown the values resulting from the updated dynamical algorithm, while in the thirdcolumn is shown the corresponding values from the previous static algorithm. Comparingwith the corresponding values from the Northern Hemisphere in Table 1 we see that the an-nual mean_edge_differences for the Southern Hemisphere all are higher by approximatelya factor 2. This is the case for both the dynamical and the static product, even though thedynamical product has slightly higher values than the static.

These annual values of mean_edge_differences in the Weddell Sea do not fulfill the accu-racy requirement of 20 km distance to the ice chart ice edge which is given in the ServiceSpecification Document [OSI SAF project team, 2014] - neither for the dynamical productnor the operational static product.

Several components can cause the higher values of mean_edge_differences in Southern

EUMETSAT OSI SAF Version 1.1 — April 2015

Page 21: Global Sea Ice Edge and Type Validation Reportosisaf.met.no/docs/osisaf_cdop2_ss2_valrep_ice-edge-type_v1p1.pdf · diance (DLI). The sea ice products include sea ice concentration,

SAF/OSI/CDOP2/MET-Norway/SCI/RP/224 17

Year Mean_edge_difference [km] Mean_edge_difference [km] Nedge

dyn stat2010 35.93 32.49 24622011 55.44 51.32 21822012 32.89 32.48 10012013 40.54 37.23 176202014 34.07 33.42 21800

Table 2: The annual mean distance between the OSI SAF ice edge and the MET-Norwayice chart edge for the Weddel Sea. Second column shows the values resulting from the dy-namical algorithm, while the third column shows the corresponding values from the previousstatic algorithm. Nedge is the total number of edge-pixels used in the estimate.

Hemisphere. First, these values only represent the summer half year. From several yearsexperience of similar validation from the Northern Hemisphere the summer values are oftena factor 2 higher or more relative to the rest of the year, see e.g. Figure 4 or the Half-YearlyOperations reports [OSI SAF project team, 2015]. Taking this difference between summermean and annual mean into account, we can expect that the Southern Hemisphere productis much closer to the target accuracy requirement than Table 2 indicates. Secondly, theregion of Weddell Sea ice chart has less good coverage of high resolution satellite data,predominantly Sentinel-1, since the region does not have high accuracy priority for receivingdata. The Weddel Sea ice chart is produced once a week and is based on a mosaic of satel-lite data within the last couple of days (with highest priority on the most recent data). Due toa larger time span in the input data the ice chart are drawn with more coarse polygons thanused in Northern Hemisphere ice charts [H. Larsen, Norwegian Ice Service, pers. comm.,April 13th 2015].

Figure 8 shows the monthly time series of the pixel-by-pixel comparison [%] between OSI SAFsea ice edge and MET-Norway ice charts covering Weddell Sea region, and Figure 9 showsthe bi-weekly time series of the pixel-by-pixel comparison against NIC ice charts on theSouthern Hemisphere. Figure 9a is the validation for the reprocessed ice edge productbased on dynamical PDF’s and Figure 9b is taken from the Half-Yearly Operations Report(2nd half 2014) [OSI SAF project team, 2015, Fig.53] and is based on static PDF’s in theoperational algorithm. The pixel-by-pixel comparison shows best performance during au-tumn, winter and spring, and – as for the Northern Hemisphere – there is a bias toward anunderestimation of the ice chart ice edge.

Note, that since the end of 2013 the number of pixels available for the comparison has beensignificantly higher (Figure 8d) and the NIC ice chart has been produced more often than bi-weekly (Figure 9). Contemporaneous, the Weddell Sea ice chart match (Figure 8a) showshigher and less variable magnitudes than earlier in the time series. In addition, the two terms,underestimate and overestimate (Figures 8b and 8c), are more comparable in magnitude,reducing the bias toward an OSI SAF-underestimation of the ice edge. These changes inthe validation values are however not seen in the time series in Figure 9, where the seasonalmatch-values seem to be stable throughout the time series and the underestimate-valueshave summer peaks of same magnitude than earlier in the time series.

Both in Figure 8 and Figure 9 we see very comparable magnitudes between the static- andthe dynamical product.

EUMETSAT OSI SAF Version 1.1 — April 2015

Page 22: Global Sea Ice Edge and Type Validation Reportosisaf.met.no/docs/osisaf_cdop2_ss2_valrep_ice-edge-type_v1p1.pdf · diance (DLI). The sea ice products include sea ice concentration,

SAF/OSI/CDOP2/MET-Norway/SCI/RP/224 18

(a) Match [%] (see Equation 3.1).

(b) Underestimate [%] (see Equation 3.2).

(c) Overestimate [%] (see Equation 3.3).

(d) N .

Figure 8: The pixel-by-pixel validation of the OSI SAF sea ice edge product against MET-Norway ice charts in the terms match, underestimate and overestimate which are defined inSection 3.1. N is the total number of pixels used in the comparison.

EUMETSAT OSI SAF Version 1.1 — April 2015

Page 23: Global Sea Ice Edge and Type Validation Reportosisaf.met.no/docs/osisaf_cdop2_ss2_valrep_ice-edge-type_v1p1.pdf · diance (DLI). The sea ice products include sea ice concentration,

SAF/OSI/CDOP2/MET-Norway/SCI/RP/224 19

(a) Validation based on dynamical PDF’s for 2010-2014.

(b) Validation based on static PDF’s back to 2005 [OSI SAF project team, 2015, Fig.53].

Figure 9: The pixel-by-pixel validation of the NH OSI SAF sea ice edge product against DMIice charts around Greenland in the terms match (light gray), underestimate (dark gray) andoverestimate (black), which are defined in Section 3.1.

EUMETSAT OSI SAF Version 1.1 — April 2015

Page 24: Global Sea Ice Edge and Type Validation Reportosisaf.met.no/docs/osisaf_cdop2_ss2_valrep_ice-edge-type_v1p1.pdf · diance (DLI). The sea ice products include sea ice concentration,

SAF/OSI/CDOP2/MET-Norway/SCI/RP/224 20

4.2 Validation of sea ice type

4.2.1 Northern Hemisphere

The time series of the multi-year ice extent [km2] for the years 2010-2015 is shown in Fig-ure 10. In Figure 10a the extent is given for the upgraded dynamical algorithm, and inFigure 10b the corresponding ice extent is given for the static algorithm included here forcomparison. The two algorithms seem to capture the same main features of the multi-yearice coverage with multi-year ice extent varying around 2.000.000 km2 and with a maximumextent in autumn 2013. A maximum extent in multi-year ice has also been reported by theNational Snow and Ice Data Center, see e.g. news on www: Reporting Climate Science[2014]. The linear trend in autumn 2014 in Figure 10a is due to missing data when repro-cessing the ice edge product with dynamical PDF’s.

(a) Multi-year ice based on dynamical PDF’s.

(b) Multi-year ice based on static PDF’s.

Figure 10: Time series of the total coverage [km2] of multi-year ice in the Northern Hemi-sphere. Black dots are daily values. Red line is the 11-days running mean, and the orangeshaded regions indicate the 11-days running standard deviation. Grey columns indicatesummer periods (the period from May to October) where OSI SAF classifies ice type asambiguous.

The sea ice type monitoring is summarized in Figure 11 which shows the monthly standarddeviation (std) of the daily variability of the multi-year ice extent, (a) for the dynamical productand (b) for the static product. All monthly std-values are within the accuracy requirement of100.000 km2, see Section 3.2.2.

EUMETSAT OSI SAF Version 1.1 — April 2015

Page 25: Global Sea Ice Edge and Type Validation Reportosisaf.met.no/docs/osisaf_cdop2_ss2_valrep_ice-edge-type_v1p1.pdf · diance (DLI). The sea ice products include sea ice concentration,

SAF/OSI/CDOP2/MET-Norway/SCI/RP/224 21

(a) Multi-year ice based on dynamical PDF’s.

(b) Multi-year ice based on static PDF’s.

Figure 11: Monthly variability (October-April) of total coverage [km2] of multi-year ice inthe Northern Hemisphere, 2010-2014. Summer periods (the period from mid-May to mid-October) are not included since OSI SAF classifies ice type as ambiguous during summer.The dashed black line represents the 100.000 km2 accuracy requirement.

EUMETSAT OSI SAF Version 1.1 — April 2015

Page 26: Global Sea Ice Edge and Type Validation Reportosisaf.met.no/docs/osisaf_cdop2_ss2_valrep_ice-edge-type_v1p1.pdf · diance (DLI). The sea ice products include sea ice concentration,

SAF/OSI/CDOP2/MET-Norway/SCI/RP/224 22

5. Conclusion

The operational OSI SAF sea ice edge and sea ice type products are based on a probabilistic(Bayesian) approach to combine ASCAT and SSMIS sensor data. At present, the algorithmis tuned to a set of static, monthly PDF’s, that were determined in 2007-2008. The proposedupgrade of these two products includes the use of dynamical PDF’s, that are updated on adaily basis from training data.

The figures and discussion in the previous chapters show how the use of dynamical PDF’sin the sea ice edge and sea ice type multi sensor product perform, and how the resultscompare with using static PDF’s.

• Sea ice type

– Northern Hemisphere The monitoring of the sea ice type product is very similarfor the dynamical product and the static product, both regarding the total coverageestimate of the multi-year ice and its monthly variability. The performance of boththe dynamical product and the static product for sea ice type are within the targetaccuracy requirements of the product.

• Sea ice edge

– Northern Hemisphere The performance of the dynamical product for sea iceedge shows results that are within the target accuracy requirements of the prod-uct. When the results for the dynamical PDF’s are compared with the perfor-mance of the static product, the results show that the dynamical product givebetter results in some cases, and worse results in other cases. On average, thestatic product gives a slightly higher score.

– Southern Hemisphere The dynamical product and the static product performvery similar for the sea ice edge on Southern Hemisphere, with the static productdoing slightly better. When compared with ice charts for summer months, theperformance of both the dynamical and the static products for sea ice edge showsresults that are not within the target accuracy requirements of the product. Buthere we compare a validation based on only summer months with an annualtarget accuracy requirement. The OSI SAF ice products normally perform worsein summer. The Weddell Sea ice chart is a new validation source for the OSI SAFsea ice products and we are still working on how to improve the validation. Alsowe are looking for other validation sources from Southern Hemisphere that wecan validate our ice product against.

Both for the Northern Hemisphere and Southern Hemisphere the mean_edge_differencesvalues are often slightly higher for the dynamical products than for the operational staticproducts, whereas the pixel-by-pixel comparison indicates similar validation results betweenthe two products. All together we believe that the updated algorithm with dynamical PDF’sperform as good as the static algorithm.

In addition, the dynamical PDF’s have important benefits compared to the static PDF’s. Es-pecially, dynamical PDF’s makes it much easier to incorporate new satellites missions in the

EUMETSAT OSI SAF Version 1.1 — April 2015

Page 27: Global Sea Ice Edge and Type Validation Reportosisaf.met.no/docs/osisaf_cdop2_ss2_valrep_ice-edge-type_v1p1.pdf · diance (DLI). The sea ice products include sea ice concentration,

SAF/OSI/CDOP2/MET-Norway/SCI/RP/224 23

production chains, as less algorithm calibration will be necessary. The static PDF’s approachrequires at least a full year of calibration before a production chain can run. Secondly, thedynamical PDF’s will quickly adapt to calibration issues that might turn up during a satelliteslife cycle. Recently, the OSI SAF sea ice concentration product was also updated to usedynamical tie-points instead of static ones.

With these results and benefits in mind, the OSI SAF project team proposes to start produc-ing the operational ice edge and type products with the dynamical PDF chain.

EUMETSAT OSI SAF Version 1.1 — April 2015

Page 28: Global Sea Ice Edge and Type Validation Reportosisaf.met.no/docs/osisaf_cdop2_ss2_valrep_ice-edge-type_v1p1.pdf · diance (DLI). The sea ice products include sea ice concentration,

SAF/OSI/CDOP2/MET-Norway/SCI/RP/224 24

References

Aaboe, S., Breivik, L.-A., and Eastwood, S. (2015a). Algorithm Theoretical Basis Docu-ment for the OSI SAF Global Sea Ice Edge (OSI-402-b) and Type (OSI-403-b) product– v1.2. Technical Report SAF/OSI/CDOP2/MET-Norway/SCI/MA/208, EUMETSAT OSISAF – Ocean and Sea Ice Sattelite Application Facility. http://osisaf.met.no/docs/.

Aaboe, S., Breivik, L.-A., Sørensen, A., Eastwood, S., and Lavergne, T. (2015b). ProductUser’s Manual for the OSI SAF Global Sea Ice Edge (OSI-402-b) and Type (OSI-402-b)product – v1.1. EUMETSAT OSI SAF – Ocean and Sea Ice Sattelite Application Facility.http://osisaf.met.no/docs/.

OSI SAF project team (2014). OSI SAF CDOP-2 Service Specification Document– v2.3. EUMETSAT OSI SAF – Ocean and Sea Ice Sattelite Application Facility.SAF/OSI/CDOP2/M-F/MGT/PL/2-003. http://www.osi-saf.org/.

OSI SAF project team (2015). OSI SAF CDOP-2 Half-Yearly Operations Report - 2nd half2014 – v1.0. Technical Report SAF/OSI/CDOP2/M-F/TEC/RP/334, EUMETSAT OSI SAF– Ocean and Sea Ice Sattelite Application Facility. http://www.osi-saf.org/.

www: Reporting Climate Science (2014). http://www.reportingclimatescience.com/news-stories/article/latest-data-shows-arctic-ice-volume-has-increased.htm.

EUMETSAT OSI SAF Version 1.1 — April 2015