PRODUCT USER MANUAL For Sea Level SLA products

Transcription

1 PRODUCT USER MANUAL For Sea Level SLA products SEALEVEL_GLO_SLA_L3_NRT_OBSERVATIONS_008_017 SEALEVEL_GLO_SLA_L3_REP_OBSERVATIONS_008_018 SEALEVEL_MED_SLA_L3_NRT_OBSERVATIONS_008_019 SEALEVEL_MED_SLA_L3_REP_OBSERVATIONS_008_020 SEALEVEL_BS_SLA_L3_NRT_OBSERVATIONS_008_022 SEALEVEL_BS_SLA_L3_REP_OBSERVATIONS_008_023 SEALEVEL_EUR_SLA_L3_NRT_OBSERVATIONS_008_024 SEALEVEL_ARC_SLA_L3_NRT_OBSERVATIONS_008_025 SEALEVEL_GLO_SLA_MAP_L4_NRT_OBSERVATIONS_008_026 SEALEVEL_GLO_SLA_MAP_L4_REP_OBSERVATIONS_008_027 SEALEVEL_MED_SLA_MAP_L4_NRT_OBSERVATIONS_008_028 SEALEVEL_MED_SLA_MAP_L4_REP_OBSERVATIONS_008_029 SEALEVEL_BS_SLA_MAP_L4_NRT_OBSERVATIONS_008_030 SEALEVEL_BS_SLA_MAP_L4_REP_OBSERVATIONS_008_031 SEALEVEL_GLO_NOISE_L4_NRT_OBSERVATIONS_008_032 SEA_LEVEL_GLO_NOISE_L4_REP_OBSERVATIONS_008_033 SEALEVEL_GLO_REF20YTO7Y_L4_OBSERVATIONS_008_034 SEALEVEL_MED_REF20YTO7Y_L4_OBSERVATIONS_008_035 SEALEVEL_BS_REF20YTO7Y_L4_OBSERVATIONS_008_036 Issue: 1.10 Contributors: Françoise Mertz (CLS), Vinca Rosmorduc (CLS), Caroline Maheu (CLS), Yannice Faugère (CLS) CMEMS version scope : Version 1.0 Approval Date : July

2 CHANGE RECORD Issue Date Description of change Author Validated by /01/15 all First version of document for V4 products F. Mertz, V. Rosmorduc, C. Maheu /03/24 all Corrections from Top level F. Mertz, V. Rosmorduc, C. Maheu /05/13 all Version 4.1, addition of HY-2A F. Mertz, V. Rosmorduc, C. Maheu /06/19 all Version 4.1, review from top level F. Mertz, V. Rosmorduc, C. Maheu /08/19 all Version September 2014 F. Mertz, V. Rosmorduc, C. Maheu /10/27 all HY-2A in REP products F. Mertz, V. Rosmorduc, C. Maheu /05/01 all Change format to fit CMEMS graphical rules /05/18 all New orbit standards GDR-E (Jason-2, Cryosat-2, AltiKa) /06/30 all Implementation of the Saral/AltiKa geodetic orbit F. Mertz, V. Rosmorduc, C. Maheu F. Mertz, V. Rosmorduc, C. Maheu G. Larnicol G. Larnicol G. Larnicol G. Larnicol G. Larnicol G. Larnicol G. Larnicol G. Larnicol F. Mertz, C. Maheu G. Larnicol /07/09 all Modifications after review F. Mertz, C. Maheu G. Larnicol /07/09 all Addition of gridded products F. Mertz, C. Maheu G. Larnicol c EU Copernicus Marine Service - Public Page 1

3 CONTENTS I. Introduction 6 II. Altimetry Principle 8 II.1. Definitions II.2. How to change the reference period III.Deactivation of the HY-2A mission (August 2015 for NRT products) 10 IV. Geodetic orbit for AltiKa (April-August 2015) 11 V. SL-TAC system 12 V.1. Introduction V.2. Near Real Time processing steps V.2.1. Input data, models and corrections applied V.2.2. Acquisition V.2.3. Homogenization V.2.4. Input data quality control V.2.5. Multi-mission cross-calibration V.2.6. Product generation V Change of Reference V Computation of raw SLA V Cross validation V Filtering and sub-sampling V.2.7. Generation of gridded Sea Level Anomalies (MSLA) products V Merging process V.2.8. Quality control V Final quality Control V Performance indicators V.3. Delayed Time processing steps V.3.1. Input data, models and corrections applied V.3.2. Acquisition V.3.3. Homogenization V.3.4. Input data quality control V.3.5. Multi-mission cross-calibration V.3.6. Product generation V Change of Reference V Computation of raw SLA V Cross validation V Filtering and sub-sampling V.3.7. Generation of gridded Sea Level Anomalies (MSLA) products V Merging process V.3.8. Quality control V Final quality Control c EU Copernicus Marine Service - Public Page 2

4 VI. COPERNICUS SL-TAC Products 39 VI.1. Near Real Time Products VI.1.1. Delay of the products VI.1.2. Temporal availibility VI.2. Delayed Time Products VI.2.1. Delay of the products VI.2.2. Temporal availibility VII.Description of the product specification 46 VII.1.General information VII.1.1.Along-track Sea Level Anomalies VII.1.2.Gridded Sea Level Anomalies VII.1.3.Gridded Noise on Sea Level Anomalies VIII. Nomenclature of files 57 VIII.1.Nomenclature of files downloaded through the CMEMS Web Portal download Service VIII.1.1.Nomenclature of the Along Track Sea Level Anomalies VIII Nomenclature of the datasets VIII Nomenclature of the NetCdf files VIII.1.2.Nomenclature of the Gridded Sea Level Anomalies VIII Nomenclature of the datasets VIII Nomenclature of the NetCdf files VIII.1.3.Nomenclature of the Gridded noise of Sea Level Anomalies VIII Nomenclature of the datasets VIII Nomenclature of the NetCdf files VIII.1.4.Nomenclature of the Gridded change of reference products VIII Nomenclature of the datasets VIII Nomenclature of the NetCdf files IX. Data format 63 IX.1. NetCdf IX.2. Structure and semantic of NetCDF along-track (L3) files IX.3. Structure and semantic of NetCDF maps (L4) files IX.4. Structure and semantic of NetCDF Gridded Noise on Sea Level Anomaly files IX.5. Structure and semantic of NetCDF Gridded Change reference files X. How to download a product 74 X.1. Download a product through the CMEMS Web Portal Directgetfile Service X.2. Download a product through the CMEMS Web Portal Ftp Service X.3. Download a product through the CMEMS Web Portal Subsetter Service XI. News and Updates 75 XI.1. [Duacs] Operational news XI.2. Updates c EU Copernicus Marine Service - Public Page 3

5 LIST OF TABLES 1 SL-TAC Near-Real Time Input data overview Corrections and models applied in SL-TAC NRT products produced from IGDRs Corrections and models applied in SL-TAC NRT products produced from OGDRs Near-Real Time Filtering and sub-sampling values Measurement noise error before/ after spatial filtering for Mediterranean and Black Sea products (cms rms) Corrections and models applied to SL-TAC DT products for TOPEX/Poseidon, Jason-1, Jason Corrections and models applied in SL-TAC DT products for ERS-1, ERS-2, Envisat Corrections and models applied in SL-TAC DT products for Cryosat-2, GFO, AltiKa and HY-2A Delayed Time Filtering and sub-sampling values Measurement noise error before/ after spatial filtering for Mediterranean and Black Sea products (cms rms) List of the time varying products in NRT List of the time invariant products in NRT List of the time varying products in Delayed Time List of the time invariant products in Delayed Time LIST OF FIGURES 1 FIGPRINCIPE SL-TAC processing sequences Overview of the near real time system data flow management Merging pertinent information from IGDR and OGDR processing Impact in cm of the reference change from 7 years to 20 years ([ ] to [ ]) 20 6 Noise Level in 1hz Jason-2 SLA estimated from mean wavenumber spectra over 2011 in 10 x10 boxes (cm rms) before 65km filtering (left) and after (right) Example with the key performance indicator on 2009/06/ Impact in cm of the reference change from 7 years to 20 years ([ ] to [ ]) 33 9 Noise Level in 1hz Jason-2 SLA estimated from mean wavenumber spectra over 2011 in 10 x10 boxes (cm rms) before 65km filtering (left) and after (right) Three merged maps are produced daily: final map (d-6), intermediate map (d-3) and preliminary map (d0) List of satellites in all-sat-merged products c EU Copernicus Marine Service - Public Page 4

6 LIST OF ACRONYMS AL AltiKa ATP Along-Track Product ADT Absolute Dynamic Topography AVISO Archiving, Validation and Interpretation of Satellite Oceanographic data BGLO Biais Grande Longueur d Onde Cal/Val Calibration - Validation CERSAT Centre ERS d Archivage et de Traitement CMA Centre Multimission Altimetry center CORSSH CORrected Sea Surface Height C2 Cryosat-2 DAC Dynamic Atmospheric Correction DT Delayed Time DTU Mean Sea Surface computed by Technical University of Danemark DUACS Data Unification and Altimeter Combination System E1 ERS-1 E2 ERS-2 EN Envisat ENN Envisat on its non repetitive orbit (since cycle 94) ECMWF European Centre for Medium-range Weather Forecasting ENACT ENhanced ocean data Assimilation and Climate prediction G2 Geosat Follow On GIM Global Ionospheric Maps GDR Geophysical Data Record(s) HY-2A Haiyang-2A IERS International Earth Rotation Service IGDR Interim Geophysical Data Record(s) J1 Jason-1 J1N Jason-1 on its new orbit (since cycle 262) J1G Jason-1 on its geodetic orbit (since May 2012) J2 Jason-2 JPL Jet Propulsion Laboratory LAS Live Access Server LWE Long Wavelength Errors MADT Map of Absolute Dynamic Topgraphy MDT Mean Dynamic Topography MOE Medium Orbit Ephemeris MP Mean Profile MSLA Map of Sea Level Anomaly MSS Mean Sea Surface NRT Near-Real Time OE Orbit Error OER Orbit Error Reduction Opendap Open-source Project for a Network Data Access Protocol PF Polynom Fit PO.DAAC Physical Oceanography Distributed Active Archive Centre c EU Copernicus Marine Service - Public Page 5

7 POE Precise Orbit Ephemeris RD Reference Document SAD Static Auxiliary Data SARAL Satellite with ARgos and ALtika SI Signed Integer SLA Sea Level Anomaly SL TAC Sea Level Thematic Assembly Centre SSALTO Ssalto multimission ground segment SSH Sea Surface Height TAC Thematic Assembly Centre T/P Topex/Poseidon TPN Topex/Poseidon on its new orbit (since cycle 369) c EU Copernicus Marine Service - Public Page 6

8 I. INTRODUCTION The Sea Level TAC (Thematic Assembly Centre) is one of the five TAC of the Copernicus Marine Environment Monitoring Service (CMEMS) project. The aim of this document is to describe the products delivered by the Sea Level TAC. The data produced in the frame of this TAC are generated by the processing system including data from all altimeter missions: HY-2A, Saral/AltiKa, Cryosat-2, OSTM/Jason-2, Jason-1, Topex/Poseidon, Envisat, GFO, ERS-1&2. The products described in this user manual are the following: 1) SEALEVEL_GLO_SLA_L3_NRT_OBSERVATIONS_008_017 SEALEVEL_MED_SLA_L3_NRT_OBSERVATIONS_008_019 SEALEVEL_BS_SLA_L3_NRT_OBSERVATIONS_008_022 SEALEVEL_EUR_SLA_L3_NRT_OBSERVATIONS_008_024 SEALEVEL_ARC_SLA_L3_NRT_OBSERVATIONS_008_025 and SEALEVEL_GLO_SLA_L3_REP_OBSERVATIONS_008_018 SEALEVEL_MED_SLA_L3_REP_OBSERVATIONS_008_020 SEALEVEL_BS_SLA_L3_REP_OBSERVATIONS_008_023 are Sea Level Anomalies observations from the altimeters. The data provided to users have a global coverage (SEALEVEL_GLO_*_OBSERVATIONS_008_*) and regional products are also computed over specific areas: Mediterranean Sea (SEALEVEL_MED_*_OBSERVATIONS_008_*) and Black Sea (SEALEVEL_BS_*_OBSERVATIONS_008_*). The following two regional products are available only in Near Real time: Europe (SEALEVEL_EUR_SLA_L3_NRT_OBSERVATIONS_008_024) and Arctic (SEALEVEL_ARC_SLA_L3_NRT_OBSERVATIONS_008_025). 2) SEALEVEL_GLO_SLA_MAP_L4_NRT_OBSERVATIONS_008_026 SEALEVEL_MED_SLA_MAP_L4_NRT_OBSERVATIONS_008_028 SEALEVEL_BS_SLA_MAP_L4_NRT_OBSERVATIONS_008_030 and SEALEVEL_GLO_SLA_MAP_L4_REP_OBSERVATIONS_008_027 SEALEVEL_MED_SLA_MAP_L4_REP_OBSERVATIONS_008_029 SEALEVEL_BS_SLA_MAP_L4_REP_OBSERVATIONS_008_031 are merged Maps of Sea Level Anomalies observations from the altimeters. 3) SEALEVEL_GLO_NOISE_L4_NRT_OBSERVATIONS_008_032 SEALEVEL_GLO_NOISE_L4_REP_OBSERVATIONS_008_033 are gridded products containing the noise of filtering of SLA Global Ocean products and are described in section V for NRT and V for REP. For Mediterranean and Black seas, only one value has been calculated over the area and is given in the same sections. c EU Copernicus Marine Service - Public Page 7

9 4) SEALEVEL_GLO_REF20YTO7Y_L4_OBSERVATIONS_008_034 SEALEVEL_MED_REF20YTO7Y_L4_OBSERVATIONS_008_035 SEALEVEL_BS_REF20YTO7Y_L4_OBSERVATIONS_008_036 are gridded products containg values to apply in order to calculate SLA with the 7 years reference period. Indeed, the products delivered since MyOcean V4 are referenced to 20 years of data whereas it was 7 years in the previous versions. Those products are thus delivered allowing the users to have time to adapt their tools to the new reference period. See sections V and V c EU Copernicus Marine Service - Public Page 8

10 II. ALTIMETRY PRINCIPLE II.1. Definitions The Altimetry gives access to the Sea Surface Height (SSH) above the reference ellipsoïd (see figure 1) SSH = Orbit - Altimetric Range The Mean Sea Surface (MSS) is the temporal mean of the SSH over a period N. It is a mean surface above the ellipsoïd and it includes the Geoid. See the detailed computation in V and V MSS N =<SSH> N The dynamical part of the signal: Sea Level Anomaly (SLA) is deduced from the SSH using a Mean Sea Surface (MSS): SLA N = SSH - MSS N Figure 1: Altimetry Principle The Mean Dynamic Topography (MDT) is the temporal mean of the SSH above the Geoid over a period N. MDT N = MSS N - Geoid The dynamical part of the absolute signal: Absolute Dynamic Topography (ADT) is deduced from the SLA using a Mean Dynamic Topography (MDT): ADT= SLA N - MDT N = SSH - MSS N - MDT N c EU Copernicus Marine Service - Public Page 9

11 II.2. How to change the reference period The delivered products have a reference period of 20 years (N). If you need to change to another reference period (P), you have to compute: Where <> X is the temporal mean over the period X SLA P = SLA N - <SLA N > P MSS P = MSS N + <SLA N > P MDT P = MDT N + <SLA N > P By definition ADT is not dependant of any reference period: ADT = MDT N + SLA N = MDT P + SLA P c EU Copernicus Marine Service - Public Page 10

12 III. DEACTIVATION OF THE HY-2A MISSION (AUGUST 2015 FOR NRT PRODUCTS) Between July 6th and 28th 2015, a first deactivation of HY-2A mission had been done because the input data were degraded and it strongly impacted the products. At the end of August 2015, the input data were again degraded and on August 28th 2015, the mission has been deactivated in the processing for the same reason as before. This time, it is planned to reintegrate HY-2A mission when the input data have a more robust validation. c EU Copernicus Marine Service - Public Page 11

13 IV. GEODETIC ORBIT FOR ALTIKA (APRIL-AUGUST 2015) The SARAL/AltiKa s ground track has been drifting from its nominal track between April 1 st, and August 11 th, Indeed, due to reaction wheel issues, SARAL station keeping maneuvers couldn t be performed nominally since March 31 st, As a result, SARAL/AltiKa s ground track has been drifting with deviations from the nominal track overtaking 10km depending on the latitude, instead of +/-1km usually. Since August 11th and thanks to many maneuvers, the platform is again under control and the nominal track has been reached. This drift has an impact on the SL-TAC AL products between April 1 st, 2015 and August 11 th, The processing for a repetitive mission (like Jason-2) includes the projection of the measurement into a theoretical track position in order to benefit from the precise Mean Profile estimated from past missions (see V and V ). For SARAL/AltiKa during April 1 st and August 11 th, as the distance between the theoretical track and the true track position is large, this projection processing induces an additional error in the product. Since mid may 2015, we have indeed observed an increase of the variability of the SARAL/AltiKa SLA at short wavelengths (< about 200km). In order to limit the degradation of the quality of the product in NRT processing, the SARAL/AltiKa mission is processed since June 30 th, 2015 as a geodetic mission, i.e. avoiding projection on a theoretical track, as currently done for Cryosat-2 or HY-2A processing. This implies that the measurement positions change from one cycle to another one. Moreover, a short wavelength error is induced on the raw SARAL/AltiKa measurement, due to the use of a gridded Mean Sea Surface solution rather than a more precise Mean Profile for SLA computation. However, SL-TAC processing includes an along-track filtering that strongly reduces this error signature on filtered products (see V and V ). As the orbit position constraint is still uncertain at this date, the NRT processing will continue as a geodetic mission. Note that in Delayed Time, the processing is adapted to the type of SARAL/AltiKa orbit: the mission is processed as a geodetic mission between March 31 st and August 11 th 2015 and processed as a repetitive mission otherwise. c EU Copernicus Marine Service - Public Page 12

14 V. SL-TAC SYSTEM V.1. Introduction This chapter presents the input data used by the system and an overview of the different processing steps necessary to produce the output data. SL-TAC system is made of two components: a Near Real Time one (NRT) and a Delayed-Time (REP) one. In NRT, the system s primary objective is to provide operational applications with directly usable high quality altimeter data from all missions available. In REP, it is to maintain a consistent and user-friendly altimeter database using the state-of-the-art recommendations from the altimetry community. Following figure gives an overview of the system, where processing sequences can be divided into 7 main steps: acquisition homogenization input data quality control multi-mission cross-calibration product generation merging final quality control. c EU Copernicus Marine Service - Public Page 13

15 Figure 2: SL-TAC processing sequences c EU Copernicus Marine Service - Public Page 14

16 V.2. Near Real Time processing steps V.2.1. Input data, models and corrections applied To produce along-track (L3) products in near-real time, the SL-TAC system uses two flows, based on the same instrumental measurements but with a different quality: The IGDRs that are the latest high-quality altimeter data produced in near-real-time. The OGDRs that include real time data (SARAL/AltiKa, OSTM/Jason-2 and Cryosat-2) to complete IGDRs. These fast delivery products do not always benefit from precise orbit determination, nor from some external model-based corrections (Dynamic Atmospheric Correction (DAC), Global Ionospheric Maps (GIM)). Integration of OGDR data increased the resilience and precision of the system. A better restitution of ocean variability is observed, especially in high energetic areas. Altimetric product Source Availability Type of orbit Jason-2 IGDR CNES ~24h CNES MOE GDR-E OGDR NOAA/EUMETSAT ~3 to 5 h Cryosat-2 IGDR-like ESA/CNES ~48 h CNES MOE GDR-E OGDR-like ESA/CNES best effort Saral/AltiKa IGDR CNES ~48 h CNES MOE GDR-D until June 30, 2015 CNES MOE GDR-E afterwards OGDR ISRO/EUMETSAT ~3 to 5 h HY-2A IGDR CNES/NSOAS best effort (with a mean delay of 72 h) CNES MOE GDR-D Table 1: SL-TAC Near-Real Time Input data overview See Figure 3: Overview of the near real time system data flow management. c EU Copernicus Marine Service - Public Page 15

17 NRT product from IGDR 1 j2 c2 al h2 Product version D CPP [4] version T patch 2 HPP [51] Orbit CNES MOE GDR-E CNES MOE GDR-E CNES MOE GDR-D until June , CNES MOE GDR-E afterwards Ionopheric Dry tropo Wet troposphere DAC Ocean tide bi-frequency altimeter range measurements GIM model (Iijima et al,1998[31]) Model computed from ECMWF Gaussian grids (S1 and S2 atm tides applied) AMR radiometer (enhancement in coastal regions) Model computed from ECMWF Gaussian grids MOG2D High Resolution forced with ECMWF pressure and wind fields (S1 and S2 were excluded) (Carrere and Lyard, 2003[6])+ inverse barometer. Filtering temporal window is recentered using forecasts GOT4v8 (S1 and S2 are included) and TPX 07.2 [22] for Arctic products Pole tide [Wahr, 1985 [69]] Solid tide earth Loading tide Sea bias state Orbit error Elastic response to tidal potential [Cartwright and Tayler, 1971[8]], [Cartwright and Edden, 1973[9]] Non Parametric SSB [N. Tran et al., 2012[65]] (with cycles J using GDR-D GOT4v8 (S1 and S2 are included) Non parametric SSB from J1 with unbiased sigma0 Hybrid SSB (method from Scharroo et al, 2004 [63] applied to al) CNES MOE GDR-D Model computed from ECMWF Gaussian grids Calculated from HPP [51]: -3.45% of SWH Global multi-mission crossover minimization (Le Traon and Ogor, 1998[40]) LW errors Optimal Interpolation [Le Traon et al., 1998[39]] Intercalibration Reference from cycle 20 Mean profile (see V and V ) Computed with 20 years of TP/J1/J2 data; referenced [1993,2012] MSS_CNES_CLS11 [48] referenced [1993,2012] Until June 30th, 2015: Computed with 15 years of E1/E2/EN data; referenced [1993,2012]; Since July 1st, 2015: MSS_CNES_CLS11 [48] referenced [1993,2012] MSS_CNES_CLS11 [48] referenced [1993,2012] (1) A flag included in the along-track files indicates the source of the production (OGDR/FDGDR or IGDR). If flag=0, the processed data comes from OGDRs/FDGDRs; if flag=1, the processed data comes from IGDRs. Table 2: Corrections and models applied in SL-TAC NRT products produced from IGDRs. c EU Copernicus Marine Service - Public Page 16

18 Product standard ref Orbit Ionopheric Dry troposphere Wet troposphere DAC Ocean tide NRT product from OGDR 2 j2 c2 al version D CPP [4] version T patch 2 bi-frequency altimeter range measurements Navigator GIM model (Iijima et al,1998[31]) Model computed from ECMWF Gaussian grids (S1 and S2 atm tides applied) AMR radiometer (enhancement in coastal regions) Model computed from ECMWF Gaussian grids MOG2D High Resolution forced with ECMWF pressure and wind fields (S1 and S2 were excluded) (Carrere and Lyard, 2003[6])+ inverse barometer. Filtering temporal window is decentered using forecasts GOT4v8 (S1 and S2 are included) and TPX 07.2 [22] for Arctic products Pole tide [Wahr, 1985 [69]] Solid tide earth Loading tide Sea bias state Elastic response to tidal potential [Cartwright and Tayler, 1971[8]], [Cartwright and Edden, 1973[9]] Non Parametric SSB [N. Tran et al., 2012[65]] (with cycles J using GDR-D GOT4v8 (S1 and S2 are included) Non parametric SSB from J1 with unbiased sigma0 Hybrid SSB (method from Scharroo et al, 2004 [63] applied to al) Orbit error Specific filtering of long-wavelength signal 3 LW errors Optimal Interpolation [Le Traon et al., 1998[39]] Intercalibration Reference from cycle 20 Mean profile Computed with 20 years of MSS_CNES_CLS11 [48] Until June 30th, 2015: (see V TP/J1/J2 data; referenced referenced [1993,2012] Computed with 15 years of and V ) [1993,2012] E1/E2/EN data; referenced [1993,2012]; Since July 1st, 2015: MSS_CNES_CLS11 [48] referenced [1993,2012] (2) A flag included in the along-track files indicates the source of the production (OGDR or IGDR). If flag=0, the processed data comes from OGDRs; if flag=1, the processed data comes from IGDRs. (3) Specific data processing was applied on long wave-length signal ( V.2.3. of the user manual) Table 3: Corrections and models applied in SL-TAC NRT products produced from OGDRs. c EU Copernicus Marine Service - Public Page 17

19 V.2.2. Acquisition The acquisition process is twofold: straightforward retrieval and reformatting of altimeter data and dynamic auxiliary data (pressure and wet troposphere correction grids from ECMWF are provided by Meteo France, TEC grids from JPL, NRT MOG2D corrections,...) from external repositories. synchronisation process. To be homogenized properly, altimeter data sets require various auxiliary data. The acquisition software detects, downloads and processes incoming data as soon as they are available on remote sites (external database, FTP site). Data are split into passes if necessary. If data flows are missing or late, the synchronisation engine put unusable data in waiting queues and automatically unfreezes them upon reception of the missing auxiliary data. This processing step delivers "raw" data, that is to say data that have been divided into cycles and passes, and ordered chronologically. The acquisition step uses two different data flows in near-real time: the OGDR flow (within a few hours), and the IGDR flow (within a few days). For each OGDR input, the system checks that no equivalent IGDR entry is available in the data base before acquisition; for each IGDR input, the system checks and delete the equivalent OGDR entry in the data base. These operations aim to avoid duplicates in SL-TAC system. Figure 3: Overview of the near real time system data flow management c EU Copernicus Marine Service - Public Page 18

20 V.2.3. Homogenization The homogenization process consists of applying the most recent corrections, models and references homogeneously for all missions and recommended for altimeter products. Each mission is processed separately as its needs depend on the base input data. The list of corrections and models currently applied is provided in tables 2 and 3 for NRT data. The system includes SLA filtering to process OGDR data. The SL-TAC processing extracts from these data sets the short scales (space and time) which are useful to better describe the ocean variability in real time, and merge this information with a fair description of large scale signals provided by the multi-satellite observation in near real time (read: IGDR-based data). Finally an "hybrid" SLA is computed. Figure 4: Merging pertinent information from IGDR and OGDR processing V.2.4. Input data quality control The Input Data Quality Control is a critical process applied to guarantee that SL-TAC uses only the most accurate altimeter data. Thanks to the high quality of current missions, this process rejects a small percentage of altimeter measurements, but these erroneous data could be the cause of a significant quality loss. The quality control relies on standard raw data editing with quality flags or parameter thresholds, but also on complex data editing algorithms based on the detection of erroneous artefacts, mono and multi-mission crossover validation, and macroscopic statistics to edit out large data flows that do not meet the system s requirements. c EU Copernicus Marine Service - Public Page 19

21 V.2.5. Multi-mission cross-calibration The Multi-mission Cross-calibration process ensures that all flows from all satellites provide a consistent and accurate information. It removes any residual orbit error (OE, Le Traon and Ogor, 1998[40]), or long wavelength error (LWE, Le Traon et al., 1998[39]), as well as large scale biases and discrepancies between various data flows. This process is based on two very different algorithms: a global multi-mission crossover minimization for orbit error reduction (OER), and Optimal Interpolation (OI) for LWE. Multi-satellite crossover determination is performed on a daily basis. All altimeter fields (measurement, corrections and other fields such as bathymetry, MSS,...) are interpolated at crossover locations and dates. Crossovers are then appended to the existing crossover database as more altimeter data become available. This crossover data set is the input of the Orbit Error Reduction (OER) method. Using the precision of the reference mission orbit (TP/J1/J2), a very accurate orbit error can be estimated. This processing step does not concern OGDR data. LWE is mostly due to residual tidal, high frequency ocean signals remaining errors and residual orbit error. The OI used for LWE reduction uses precise parameters derived from: accurate statistical description of sea level variability regional correlation scales mission-specific noise and precise assumptions on the long wavelength errors to be removed (from a recent analysis of one year of data from each mission). V.2.6. Product generation The product generation process is composed of four steps: computation of raw SLA, cross-validation, filtering&sub-sampling, and generation of by-products. c EU Copernicus Marine Service - Public Page 20

22 V Change of Reference As from April 2014, an important modification has been implemented in SL-TAC products. Indeed, the period of reference was 7 years in older version (from 1993 to 1999) and it is now 20 years (from 1993 to 2012). This takes into account the variations of the oceans in the last years, as shown on Figure 5. The detailed informations can be found in the document: MYO2-SL-QUID v1.0.pdf and at the following url: r2267_9_sltac_technical_note.pdf. In order to give the users the time to adapt to this new reference period, three gridded products are delivered for global ocean, Mediterranean and Black seas: SEALEVEL_GLO_REF20YTO7Y_L4_OBSERVATIONS_008_034 SEALEVEL_MED_REF20YTO7Y_L4_OBSERVATIONS_008_035 and SEALEVEL_BS_REF20YTO7Y_L4_OBSERVATIONS_008_036 If you may wish to come back to the 7 year reference period (=SLA 7years ): for each measurement of SLA 20years from the new products, since the REF20YTO7Y products are gridded, you need to interpolate at the location of the SLA measurement the variable ref20yto7y (see IX.5. for information about the format of the file) calculate SLA 7years =SLA 20years from the new products - ref20yto7y previously calculated Figure 5: Impact in cm of the reference change from 7 years to 20 years ([ ] to [ ]) c EU Copernicus Marine Service - Public Page 21

23 V Computation of raw SLA The SSH anomalies are used in oceanographic studies. They are computed from the difference of the instantaneous SSH minus a temporal reference. This temporal reference can be a Mean Profile (MP) in the case of repeat track or a gridded MSS when the repeat track cannot be used. The errors affecting the SLAs, MPs and MSS have different magnitudes and wavelengths. The computation of the SLAs and their errors associated are detailed in Dibarboure et al, 2010 [14]. Use of a Mean Profile In the repeat track analysis at 1 Hz (when the satellites flies over a repetitive orbit), measurements are resampled along a theoretical ground track (or mean track) associated to each mission. Then a Mean Profile (MP) is subtracted from the re-sampled data to obtain SLA. The MP is a time average of similarly re-sampled data over a long period. The Mean Profile used for Saral/Altika until June 30th, 2015 is computed with 15 years of ERS-1, ERS-2 and Envisat, referenced to the period [1993, 2012]. Since July 1st, 2015, no Mean Profile can be used for AltiKa because the orbit is drifting. The MSS is used instead of the MP (see below). The Mean Profile used for Jason-2 is computed with 20 years of T/P, Jason-1 and Jason-2, referenced to years [1993, 2012]. No Mean Profile can be used for Cryosat-2 mission (c2). The MSS must be used instead (see below). No Mean Profile can be used for HY-2A mission (h2) because there are not enough data to calculate it. The MSS must be used instead (see below). Computation of a Mean Profile The computation of a Mean Profile is not a simple average of similarly co-located SSH data from the same ground track on the maximum period of time as possible. Indeed, as the satellite ground track is not perfectly controlled and is often kept in a band of about 1km wide, precise cross-track projection and/or interpolation schemes are required to avoid errors. The ocean variability is removed to minimize the seasonal/interannual aliasing effects. The mesoscale variability error is eliminated with an iterative process using a priori knowledge from Sea Level maps derived from previous iterations or from other missions.this process enables us to reference the mean profiles for all missions to a common period (reference period) for the sake of consistency with other missions. The reference period is [1993, 2012] and is thus independent from the number of years used to compute mean profiles. Moreover, the inter-annual variabilty error is accounted for by using the MSS. Finally, for these Mean Profiles, the latest standards and a maximum of data were used in order to increase as much as possible the quality of their estimation. Note that a particular care was brought to the processing near coasts. Use of a MSS The repeat track analysis is impossible for Cryosat-2 mission (c2) and AltiKa mission after July 1st, 2015 because the satellite is not in a repetitive orbit phase. Moreover, it is not possible for the moment to compute a mean profile for HY-2A because there is not enough data to compute it.. The alternative is to use the MSS instead. The gridded MSS is derived from along track MPs and data from geodetic phases. Thus any error c EU Copernicus Marine Service - Public Page 22

24 on the MP is also contained in the MSS. There are essentially 4 types of additional errors on gridded MSS which are hard to quantify separately: To ensure a global MSS coherency between all data sets, the gridding process averages all sensorspecific errors and especially geographically correlated ones. The gridding process has to perform some smoothing to make up for signals which cannot be resolved away from known track, degrading along-track content. There are also errors related to the lack of spatial and temporal data (omission errors). The error stemming from the geodetic data: the variability not properly removed before the absorption in the MSS. The MSS used in the products is MSS_CNES_CLS11 [48], referenced [1993,2012] except for the Arctic regional products, where the DTU10 MSS [2] referenced [1993,2012] is taken into account. Indeed, the accuracy of DTU10 compared to CNES_CLS11 is further demonstrated by the significant sea level anomaly variance reduction found over the whole Arctic Ocean as shown in Prandi et al., [52].. V Cross validation After the repeat track analysis, the cross-validation technique is used as the ultimate screening process to detect isolated and slightly erroneous measurements. Small SLA flows are compared to previous and independent SLA data sets using a- 12 year climatology and a 3 sigma criteria for outlier removal. c EU Copernicus Marine Service - Public Page 23

25 V Filtering and sub-sampling Residual noise and small scale signals are then removed by filtering the data using a Lanczos filter. As data are filtered from small scales, a sub-sampling is finally applied. Along-track SLA are then produced (those products are along-track *vfec* files). The filtering and sub-sampling is adapted to each region and product as a function of the characteristics of the area and of the assimilation needs. The values taken into account are the following: Filtering Sub-sampling Global 65 km 14km Mediterranean Sea about 40km 14km Black Sea about 40km 14km Europe about 35km No (7km) Arctic about 35km No (7km) Table 4: Near-Real Time Filtering and sub-sampling values Noise estimation Since April 2014, a new filtering level takes into account the mesoscale capability computed for the Jason- 2 1hz SLA. Users recommendations from the TAPAS working group motivated the choice of a spatially uniform cut-off length to filter SLA in association with a more accurate measurement error description. Therefore, a cut-off length of 65km is applied everywhere to filter SLA. Compared to the previous SLA filtering applied, it changes drastically the SLA resolution at low latitudes. In these areas, cut-off lengths are reduced from more than 100km (see detailed informations in MYO2-SL-QUID v1.0). In wavenumber spectra, the 1hz measurement error is the noise level estimated as the mean value of energy at high wavenumbers (below 20km in term of wave length). The 1hz noise level spatial distribution (Figure 6) follows the instrumental white-noise linked to the Surface Wave Height but also connections with the backscatter coefficient. The full understanding of this hump of spectral energy (Dibarboure et al., 2013 [12]) still remain to be achieved and overcome with new retracking, new editing strategy or new technology. Users of non-filtered SLA will have to use the 1hz estimated white noise (Figure 6 left) whereas users of filtered SLA the remaining error level after filtering (Figure 6 right). c EU Copernicus Marine Service - Public Page 24

26 Figure 6: Noise Level in 1hz Jason-2 SLA estimated from mean wavenumber spectra over 2011 in 10 x10 boxes (cm rms) before 65km filtering (left) and after (right) The regional products as Mediterranean Sea, Black Sea cover too small areas to be concerned by such a map of 1hz measurement error. In this case, a mean wavenumber spectrum has been computed over each area and the noise level estimated from it. A constant value is then prescribed for the 1hz noise level as well as the remaining error after filtering: Before Filtering After Filtering Cryosat Jason Saral/AltiKa HY-2A Table 5: Measurement noise error before/ after spatial filtering for Mediterranean and Black Sea products (cms rms) V.2.7. Generation of gridded Sea Level Anomalies (MSLA) products V Merging process The Merging process is twofold: mapping and generation of by-products. A mapping procedure using optimal interpolation with realistic correlation functions is applied to produce SLA maps (MSLA or L4 products) at a given date. The procedure generates a combined map merging measurements from all available altimeter missions (Ducet et al., 2000[20]). The mapping process takes into account an updated suboptimal Optimal Interpolation parameterization to minimize transition artefacts. More accurate correlation scales are used compared to Ducet et al., taking into account optimally the spatial variability of the signal. Combining data from different missions significantly improves the estimation of mesoscale signals (Le Traon and Dibarboure, 1999[41]), (Le Traon et al., 2001[42]), (Pascual et al., 2006[49]). Several improvements were made compared to the version used by (Le Traon et al., 1998[39]). An improved statistical description of sea level variability, noise and long wavelength errors is used. Covariance functions including propagation velocities that depend on geographical position were thus used. For each grid point, the c EU Copernicus Marine Service - Public Page 25

27 zonal and meridional spatial scales, the time scale and the zonal and meridional propagation velocities were adjusted from five years of TP+ERS combined maps. In addition to instrumental noise, a noise of 10% of the signal variance was used to take into account the small scale variability which cannot be mapped and should be filtered in the analysis. Time window In the NRT DUACS processing, contrary to DT case, the products cannot be computed with a centred computation time window for OER, LWE and mapping processes: indeed, as the future data are not available yet, the computation time window is not centered (for each day of production, 3 maps are computed: for the maps of date D, D-3 and D-6 are using respectively the data in the time intervall of [D-42, D], [D-3-42, D+3] and [D-6-42, D+6]). OGDR specificity SLA computation from OGDR is based on the same algorithms, only parameters are different to take into account OGDR specificity. LWE and mapping process are based on IGDR and GDR available residuals, also with specific parameters. Number of satellites to compute the maps The maps are computed with all the satellites available (up to 4 satellites) for each date. Thus it has the best possible sampling. This series is better in quality but not homogeneous over the time period, because it is based on the best sampling available in time. The figure 11 lists the missions taken into account for the computation of the maps. Note that the list of missions is also provided in the gridded file in the global attribute "platform". Formal mapping error The formal mapping error represents a purely theoretical mapping error. It mainly represents errors induced by the constellation sampling capability and consistency with the spatial/temporal scales considered, as described in Le Traon et al (1998) [39] or Ducet et al (2000) [20]. The formal mapping error is expressed in meters and is delivered in the same NetCDF file as the SLA. V.2.8. Quality control The production of homogeneous products with a high quality data and within a short delay is the key feature of the SL-TAC processing system. But some events (failure on payload or on instruments, delay, maintenance on servers), can impact the quality of measurements or the data flows. A strict quality control on each processing step is indispensable to appreciate the overall quality of the system and to provide the best user services. V Final quality Control The Quality Control is the final process used by SL-TAC before product delivery. In addition to daily automated controls and warnings to the operators, each production delivers a large QC Report composed of detailed logs, figures and statistics of each processing step. Altimetry experts analyse these reports twice a week (only for internal validation, those reports are not disseminated). V Performance indicators c EU Copernicus Marine Service - Public Page 26

28 To appreciate the quality situation of the DUACS system, performance indicators are computed daily. They aim at evaluate the status of the main processing steps of the system: the input data availability, the input data coverage, the input data quality and the output product quality. These indicators are computed for each and every currently working satellite, and combined to obtain the overall status. Figure 7: Example with the key performance indicator on 2009/06/27 See the description, the latest and previous indicators on Aviso website: ssaltoduacs-multimission-altimeter-products/key-performance-indicators.html c EU Copernicus Marine Service - Public Page 27

29 V.3. Delayed Time processing steps V.3.1. Input data, models and corrections applied Delayed Time products are generated: from Aviso GDR products for Saral/AltiKa, Jason-1 (historical, tandem track and geodetic track), Jason-2 from ESA GDR products for Cryosat-2, Envisat (historical, new orbit and drifting track), ERS-1 (historical and geodetic track) and ERS-2 from NOAA GDR for GFO. from NSOAS GDR for HY-2A. The GDR products are computed with updated orbit and corrections as described in the following table. c EU Copernicus Marine Service - Public Page 28

30 Delayed-Time product j2 j1;j1n;j1g tp;tpn Product GDR-D GDR-D GDR-C Orbit Cnes POE (GDR-D standards for cycles 253 and GDR-E afterwards) Cnes POE (GDR-D standards) GSFC (ITRF 2005,GRACE last standards) Ionopheric Dual frequency altimeter range measurements Dual frequency (Topex), DORIS (POSEIDON) Dry troposphere Model computed from Model computed from Model computed from ERA ECMWF Gaussian grids (new S1 and S2 atm. tides applied) ECMWF rectangular grids (new S1 and S2 atm. tides included) interim Gaussian grids (new S1 and S2 atm. tides included) Wet troposphere JMR/AMR radiometer TMR radiometer (Scharroo et al [62]) DAC MOG2D High Resolution forced with ECMWF pressure and wind fields (S1 and S2 excluded) + inverse barometer computed from rectangular grids MOG2D High Resolution forced with ERA-Interim pressure and wind fields (S1 and S2 excluded) + inverse barometer computed from rectangular grids Ocean tide GOT4v8 (S1 and S2 are included) Pole tide [Wahr, 1985[69]] Solid earth Elastic response to tidal potential Cartwright and Tayler, 1971[8],Cartwright and Edden, 1973[9] tide Loading tide GOT4v8 (S1 parameter is included) Sea state Non Parametric SSB [N. Tran Non Parametric SSB [N. Tran Non parametric SSB [N. Tran and bias et al., 2012[65]] (with c J using GDR-D et al., 2012[65]](with c J using GDR-C standards and GDR-D orbit al. 2010[67]] (using c with GSFC orbit for TP-A; c with GSFC orbit for TP-B) Orbit error Global multi-mission crossover minimization (Le Traon and Ogor, 1998 [40]) LW errors Optimal Interpolation (Le Traon et al., 1998[39]) Intercalibration Reference from cycle 11 Reference from cycle 1 to 354 Global Bias Calibrated on J1+TP Mean Sea Level (using c 1-11 J2); MSL zero reference in early 1993 Calibrated on TP Mean Sea Level (using c 1-11 J1); MSL zero reference in early 1993 MSL zero reference in early 1993 Regional Bias Calibration on J2 Calibration on J1+J2 Mean profile Computed with 20 years of TP/J1: Computed with 20 years of TP/J1/J2 data; referenced (see V TP/J1/J2 data; referenced [1993,2012]; and V ) [1993,2012] TPN/J1N: computed with 6 years of TPN/J1N; referenced [1993,2012] Period of use from cycle 9 from cycle 9 cycles 1 to 481 Table 6: Corrections and models applied to SL-TAC DT products for TOPEX/Poseidon, Jason-1, Jason-2. The period of reference has changed since April 2014: it is now 20 years instead of 7 years. Please refer to section V c EU Copernicus Marine Service - Public Page 29

31 DUACS DT product e1 e2 en;enn Product OPR GDR-V2.1+ Orbit Reaper (Rudenko et al., 2012[61]) Cnes POE (GDR-D standards) Ionopheric Dry troposphere Wet troposphere DAC Reaper (NIC09 model, Scharro and Smith, 2010, [64]) Bent model (c 36), GIM model (c 37) ((Iijima et al, 1998 [31]) Model computed from ERA interim Gaussian grids (new S1 and S2 atm. tides included) MWR radiometer MWR corrected for 23.6GHz TB drift (Scharroo et al., 2004[62])+ Neural Network algorithm (Tran and Obligis, 2003[66] MOG2D High Resolution forced with ERA-Interim pressure and wind fields (S1 and S2 excluded) + inverse barometer computed from rectangular grids Bi-frequency altimeter range measurement (c 64); GIM model c 65 (Iijima et al., 1999[31]) corrected for 8 mm bias Model computed from ECMWF Gaussian grids (new S1 and S2 atm. tides applied) Cycles 94: MWR (dist 50km from the coasts), ECMWF model (50 dist 10km) Cycles 95:MWR. MOG2D High Resolution forced with ECMWF pressure and wind fields (S1 and S2 excluded) + inverse barometer computed from rectangular grids Ocean tide GOT4v8 (S1 and S2 are included) Pole tide [Wahr, 1985[69]]] Solid earth Elastic response to tidal potential Cartwright and Tayler, 1971[8],Cartwright and Edden, 1973[9] tide Loading tide GOT4v8 (S1 parameter is included) Sea state BM3 [Gaspar and Ogor Non parametric SSB [Mertz et Non Parametric SSB [N. Tran et bias 1994[24]] al., 2005[47]] al., 2012[65]] compatible with enhanced MWR Orbit error Global multi-mission crossover minimization (Le Traon and Ogor, 1998 [40]) LW errors Optimal Interpolation (Le Traon et al., 1998[39]) Global Bias Calibrated on TP/J1/J2 Mean Sea Level Mean profile Computed with 15 years of E1/E2/EN data; referenced EN: Computed with 15 years of (see V [1993,2012] E1/E2/EN data; referenced and V ) [1993,2012] ENN: use of MSS_CNES_CLS11 [48], referenced [1993,2012] Period of use cycles 15 to 43 Including E-F geodetic phases cycles 1 to 83 from cycle 9 Table 7: Corrections and models applied in SL-TAC DT products for ERS-1, ERS-2, Envisat The period of reference has changed since April 2014: it is now 20 years instead of 7 years. Please refer to section V c EU Copernicus Marine Service - Public Page 30

32 DT product c2 g2 al h2 Product CPP CNES [4] GDR (NOAA) GDR-T patch2 GDR (NSOAS) Orbit CNES POE (GDR-D standards for cycles 66 and GDR-E afterwards) GSFC (ITRF 2005, GRACE last standards Cnes POE (GDR-D standards for cycles 23 and GDR-E afterwards) Cnes POE (GDR-D standards) Ionopheric From GIM model (Iijima et al, 1998 [31]) Dry troposphere Model computed from Model computed from Model computed from ECMWF Gaussian ECMWF Gaussian grids (new S1 and S2 atm. tides applied) ECMWF rectangular grids (new S1 and S2 atm. tides included) grids (new S1 and S2 atm. tides applied) Wet troposphere ECMWF Model GFO radiometer radiometer ECMWF Model DAC MOG2D High Resolution forced with ECMWF pressure and wind fields (S1 and S2 excluded) + inverse barometer computed from rectangular grids Ocean tide GOT4v8 (S1 and S2 are included) Pole tide [Wahr, 1985[69]]] Solid earth Elastic response to tidal potential Cartwright and Tayler, 1971[8],Cartwright and Edden, tide 1973[9] Loading tide GOT4v8 (S1 parameter is included) Sea bias state Non parametric SSB from J1 with unbiased sigma0 Non parametric SSB [N. Tran et al., 2010[67]] (using c 130 to 172 with GSFC orbit) Hybrid SSB (method from Scharroo et al, 2004 [63] applied to al) Orbit error Global multi-mission crossover minimization (Le Traon and Ogor, 1998 [40]) LW errors Optimal Interpolation (Le Traon et al., 1998[39]) Global Bias Calibrated on TP/J1/J2 Mean Sea Level Mean profile Use of Computed with 8 (see V MSS_CNES_CLS11 years of G2 ; and V ) [48], referenced [1993,2012] referenced [1993,2012] Until March 31st, 2015 and after August 12th 2015: Computed with 15 years of E1/E2/EN data; referenced [1993,2012]; Between April 1st and August 11th, 2015: Use of MSS_CNES_CLS11 [48], referenced [1993,2012] Linear model Use of MSS_CNES_CLS11 [48], referenced [1993,2012] Period of use from cycle 14 cycles 37 to 222 from cycle 1 from cycle 67 Table 8: Corrections and models applied in SL-TAC DT products for Cryosat-2, GFO, AltiKa and HY-2A The period of reference has changed since April 2014: it is now 20 years instead of 7 years. Please refer to section V c EU Copernicus Marine Service - Public Page 31

33 V.3.2. Acquisition The acquisition process in Delayed time consists in a synchronisation process of all the auxiliary data required to homogenize propely the altimeter data sets. The acquisition step uses the GDRs provided by the agencies. V.3.3. Homogenization The homogenization process consists of applying the most recent corrections, models and references homogeneously for all missions and recommended for altimeter products. Each mission is processed separately as its needs depend on the base input data. The list of corrections and models currently applied is provided in tables 6 for DT data. V.3.4. Input data quality control The Input Data Quality Control is a critical process applied to guarantee that SL-TAC uses only the most accurate altimeter data. Thanks to the high quality of current missions, this process rejects a small percentage of altimeter measurements, but these erroneous data could be the cause of a significant quality loss. The quality control relies on standard raw data editing with quality flags or parameter thresholds, but also on complex data editing algorithms based on the detection of erroneous artefacts, mono and multi-mission crossover validation, and macroscopic statistics to edit out large data flows that do not meet the system s requirements. c EU Copernicus Marine Service - Public Page 32

34 V.3.5. Multi-mission cross-calibration The Multi-mission Cross-calibration process ensures that all flows from all satellites provide a consistent and accurate information. It removes any residual orbit error (OE, Le Traon and Ogor, 1998[40]), or long wavelength error (LWE, Le Traon et al., 1998[39]), as well as large scale biases and discrepancies between various data flows. This process is based on two very different algorithms: a global multi-mission crossover minimization for orbit error reduction (OER), and Optimal Interpolation (OI) for LWE. Multi-satellite crossover determination is performed on a daily basis. All altimeter fields (measurement, corrections and other fields such as bathymetry, MSS,...) are interpolated at crossover locations and dates. Crossovers are then appended to the existing crossover database as more altimeter data become available. This crossover data set is the input of the Orbit Error Reduction (OER) method. Using the precision of the reference mission orbit (TP/J1/J2), a very accurate orbit error can be estimated. LWE is mostly due to residual tidal, high frequency ocean signals remaining errors and residual orbit error. The OI used for LWE reduction uses precise parameters derived from: accurate statistical description of sea level variability regional correlation scales mission-specific noise and precise assumptions on the long wavelength errors to be removed (from a recent analysis of one year of data from each mission). V.3.6. Product generation The product generation process is composed of four steps: computation of raw SLA, cross-validation, filtering&sub-sampling, and generation of by-products. c EU Copernicus Marine Service - Public Page 33

35 V Change of Reference As from April 2014, an important modification has been implemented in SL-TAC products. Indeed, the period of reference was 7 years in older version (from 1993 to 1999) and it is now 20 years (from 1993 to 2012). This takes into account the variations of the oceans in the last years, as shown on Figure 8. The detailed informations can be found in the document: MYO2-SL-QUID v1.0.pdf and at the following url: r2267_9_sltac_technical_note.pdf. In order to give the users the time to adapt to this new reference period, three gridded products are delivered for global ocean, Mediterranean and Black seas: SEALEVEL_GLO_REF20YTO7Y_L4_OBSERVATIONS_008_034 SEALEVEL_MED_REF20YTO7Y_L4_OBSERVATIONS_008_035 and SEALEVEL_BS_REF20YTO7Y_L4_OBSERVATIONS_008_036 If you may wish to come back to the 7 year reference period (=SLA 7years ): for each measurement of SLA 20years from the new products, since the REF20YTO7Y products are gridded, you need to interpolate at the location of the SLA measurement the variable ref20yto7y (see IX.5. for information about the format of the file) calculate SLA 7years =SLA 20years from the new products - ref20yto7y previously calculated Figure 8: Impact in cm of the reference change from 7 years to 20 years ([ ] to [ ]) c EU Copernicus Marine Service - Public Page 34

36 V Computation of raw SLA The SSH anomalies are used in oceanographic studies. They are computed from the difference of the instantaneous SSH minus a temporal reference. This temporal reference can be a Mean Profile (MP) in the case of repeat track or a gridded MSS when the repeat track cannot be used. The errors affecting the SLAs, MPs and MSS have different magnitudes and wavelengths. The computation of the SLAs and their errors associated are detailed in Dibarboure et al, 2010 [14]. Use of a Mean Profile In the repeat track analysis at 1 Hz (when the satellites flies over a repetitive orbit), measurements are resampled along a theoretical ground track (or mean track) associated to each mission. Then a Mean Profile (MP) is subtracted from the re-sampled data to obtain SLA. The MP is a time average of similarly re-sampled data over a long period. The Mean Profile used for Saral/Altika until March 31th, 2015 and after August 12th, 2015 is computed with 15 years of ERS-1, ERS-2 and Envisat, referenced to the period [1993, 2012]. Between April 1st and August 11th, 2015, no Mean Profile can be used for AltiKa because the orbit is drifting. The MSS is used instead of the MP (see below). The Mean Profile used for T/P (cycles 1 to 364), Jason-1 (cycles 1 to 259) and Jason-2 is computed with 20 years of T/P, Jason-1 and Jason-2, referenced to years [1993, 2012]. The Mean Profile used for T/P (cycles 368 to 481) and from Jason-1 cycle 262 to 374 (where satellites are on interleaved ground-tracks) is computed with 6 years of T/P and Jason-1, referenced to the period [1993, 2012]. The Mean Profile used for ERS-1 in its 35 days repetitive orbit mission, ERS-2, and Envisat (only for the first orbit, before November 2010) is computed with 15 years of ERS-1, ERS-2 and Envisat, referenced to the period [1993, 2012]. No Mean Profile can be used for Envisat on its new orbit (enn), Jason-1 on its geodetic orbit (j1g) nor for Cryosat-2 mission (c2). The MSS must be used instead (see below). No Mean Profile can be used for HY-2A mission (h2) because there are not enough data to calculate it. The MSS must be used instead (see below). The Mean Profile used for GFO is computed with 8 years of G2, referenced to the period [1993,2012]. Computation of a Mean Profile The computation of a Mean Profile is not a simple average of similarly co-located SSH data from the same ground track on the maximum period of time as possible. Indeed, as the satellite ground track is not perfectly controlled and is often kept in a band of about 1km wide, precise cross-track projection and/or interpolation schemes are required to avoid errors. The ocean variability is removed to minimize the seasonal/interannual aliasing effects. The mesoscale variability error is eliminated with an iterative process using a priori knowledge from Sea Level maps derived from previous iterations or from other missions.this process enables us to reference the mean profiles for all missions to a common period (reference period) for the sake of consistency with other missions. The reference period is [1993, 2012] and is thus independent from the number of years used to compute mean profiles. c EU Copernicus Marine Service - Public Page 35

37 Moreover, the inter-annual variabilty error is accounted for by using the MSS. Finally, for these Mean Profiles, the latest standards and a maximum of data were used in order to increase as much as possible the quality of their estimation (see tables 6 to 8: Corrections and models applied in SSALTO/DUACS Delayed-Time products). Note that a particular care was brought to the processing near coasts. Use of a MSS The repeat track analysis is impossible for ERS-1 for its 168 days geodetic mission (phases E-F from April 1994 to March 1995) or for Envisat since November 2010, for Jason-1 on its geodetic orbit (j1g), for Cryosat- 2 mission (c2) and AltiKa mission between April 1st and August 11th, 2015 because the satellite is not in a repetitive orbit phase. Moreover, it is not possible for the moment to compute a mean profile for HY-2A because there is not enough data to compute it.. The alternative is to use the MSS instead. The gridded MSS is derived from along track MPs and data from geodetic phases. Thus any error on the MP is also contained in the MSS. There are essentially 4 types of additional errors on gridded MSS which are hard to quantify separately: To ensure a global MSS coherency between all data sets, the gridding process averages all sensorspecific errors and especially geographically correlated ones. The gridding process has to perform some smoothing to make up for signals which cannot be resolved away from known track, degrading along-track content. There are also errors related to the lack of spatial and temporal data (omission errors). The error stemming from the geodetic data: the variability not properly removed before the absorption in the MSS. The MSS used in the products is MSS_CNES_CLS11 [48], referenced [1993,2012] V Cross validation After the repeat track analysis, the cross-validation technique is used as the ultimate screening process to detect isolated and slightly erroneous measurements. Small SLA flows are compared to previous and independent SLA data sets using a- 12 year climatology and a 3 sigma criteria for outlier removal. c EU Copernicus Marine Service - Public Page 36

38 V Filtering and sub-sampling Residual noise and small scale signals are then removed by filtering the data using a Lanczos filter. As data are filtered from small scales, a sub-sampling is finally applied. Along-track SLA are then produced (those products are along-track *vfec* files). Note that in Delayed-time, files *vxxc* are also produced: they are not filtered and not sub-sampled). The filtering and sub-sampling is adapted to each region and product as a function of the characteristics of the area and of the assimilation needs. The values taken into account for the filtering are the following: Filtering Sub-sampling Global 65 km 14km for filtered, 7km for unfiltered Mediterranean Sea about 40km 14km for filtered, 7km for unfiltered Black Sea about 40km 14km for filtered, 7km for unfiltered Table 9: Delayed Time Filtering and sub-sampling values Noise estimation Since April 2014, a new filtering level takes into account the mesoscale capability computed for the Jason- 2 1hz SLA. Users recommendations from the TAPAS working group motivated the choice of a spatially uniform cut-off length to filter SLA in association with a more accurate measurement error description. Therefore, a cut-off length of 65km is applied everywhere to filter SLA. Compared to the previous SLA filtering applied, it changes drastically the SLA resolution at low latitudes. In these areas, cut-off lengths are reduced from more than 100km (see detailed informations in MYO2-SL-QUID v1.0). In wavenumber spectra, the 1hz measurement error is the noise level estimated as the mean value of energy at high wavenumbers (below 20km in term of wave length). The 1hz noise level spatial distribution (Figure 9) follows the instrumental white-noise linked to the Surface Wave Height but also connections with the backscatter coefficient. The full understanding of this hump of spectral energy (Dibarboure et al., 2013 [12]) still remain to be achieved and overcome with new retracking, new editing strategy or new technology. Users of non-filtered SLA will have to use the 1hz estimated white noise (Figure 9 left) whereas users of filtered SLA the remaining error level after filtering (Figure 9 right). Figure 9: Noise Level in 1hz Jason-2 SLA estimated from mean wavenumber spectra over 2011 in 10 x10 boxes (cm rms) before 65km filtering (left) and after (right) c EU Copernicus Marine Service - Public Page 37

39 The regional products as Mediterranean Sea, Black Sea cover too small areas to be concerned by such a map of 1hz measurement error. In this case, a mean wavenumber spectrum has been computed over each area and the noise level estimated from it. A constant value is then prescribed for the 1hz noise level as well as the remaining error after filtering: Before Filtering After Filtering Cryosat Jason Saral/AltiKa HY-2A ERS ERS Envisat Geosat Follow-on Jason Topex/Poseidon Table 10: Measurement noise error before/ after spatial filtering for Mediterranean and Black Sea products (cms rms) V.3.7. Generation of gridded Sea Level Anomalies (MSLA) products V Merging process The Merging process is twofold: mapping and generation of by-products. A mapping procedure using optimal interpolation with realistic correlation functions is applied to produce SLA maps (MSLA or L4 products) at a given date. The procedure generates a combined map merging measurements from all available altimeter missions (Ducet et al., 2000[20]). The mapping process takes into account an updated suboptimal Optimal Interpolation parameterization to minimize transition artefacts. More accurate correlation scales are used compared to Ducet et al., taking into account optimally the spatial variability of the signal. Combining data from different missions significantly improves the estimation of mesoscale signals (Le Traon and Dibarboure, 1999[41]), (Le Traon et al., 2001[42]), (Pascual et al., 2006[49]). Several improvements were made compared to the version used by (Le Traon et al., 1998[39]). An improved statistical description of sea level variability, noise and long wavelength errors is used. Covariance functions including propagation velocities that depend on geographical position were thus used. For each grid point, the zonal and meridional spatial scales, the time scale and the zonal and meridional propagation velocities were adjusted from five years of TP+ERS combined maps. In addition to instrumental noise, a noise of 10% of the signal variance was used to take into account the small scale variability which cannot be mapped and should be filtered in the analysis. Time window In the DT processing, the products can be computed optimally with a centred computation time window for OER, LWE and mapping processes (if D is the date of the map, the time intervall of the data taken into c EU Copernicus Marine Service - Public Page 38

40 account for the computation of the map is [D-6 weeks, D+6 weeks]). Number of satellites to compute the maps The maps are computed with all the satellites available (up to 4 satellites) for each date. Thus it has the best possible sampling. This series is better in quality but not homogeneous over the time period, because it is based on the best sampling available in time. The figure 11 lists the missions taken into account for the computation of the maps. Note that the list of missions is also provided in the gridded file in the global attribute "platform". Formal mapping error The formal mapping error represents a purely theoretical mapping error. It mainly represents errors induced by the constellation sampling capability and consistency with the spatial/temporal scales considered, as described in Le Traon et al (1998) [39] or Ducet et al (2000) [20]. The formal mapping error is expressed in meters and is delivered in the same NetCDF file as the SLA. V.3.8. Quality control The production of homogeneous products with a high quality data and within a short delay is the key feature of the SL-TAC processing system. But some events (failure on payload or on instruments, delay, maintenance on servers), can impact the quality of measurements or the data flows. A strict quality control on each processing step is indispensable to appreciate the overall quality of the system and to provide the best user services. V Final quality Control The Quality Control is the final process used by SL-TAC before product delivery. In addition to daily automated controls and warnings to the operators, each production delivers a large QC Report composed of detailed logs, figures and statistics of each processing step. Altimetry experts analyse these reports twice a week (only for internal validation, those reports are not disseminated). c EU Copernicus Marine Service - Public Page 39

41 VI. COPERNICUS SL-TAC PRODUCTS VI.1. Near Real Time Products The purpose of the NRT component is the acquisition of altimeter data from various altimeter missions in near-real time (i.e. within a few days at most) and in real-time for global area on all satellites (i.e. within a few hours), the validation and correction of these altimeter data sets (i.e edition and selection, update of corrections and homogenization, orbit error reduction) in order to produce each day along-track products and gridded products. Exploitation of real time OGDR/FDGDR data allows the DUACS system to produce multi-mission maps with 0-day and 3-day delay whereas historical NRT (IGDR-based) production have a 6-day delay (induced by historical trade-off in terms of timeliness vs quality). The quality measurements in the NRT processing is more sensitive to the number of altimeter missions involved in the system. This is mainly due to the orbit error and the non-centered processing time-window (in NRT case, "future" data are not available; the computation time window takes into account only the 6 weeks before the date). If two altimeters are acknowledged as the bare minimum needed to observe mesoscale signals in DT maps, three or even four missions are needed to obtain equivalent accuracy in NRT (Pascual et al., 2006[49]). As described in V and V time invariant products are also disseminated: SEALEVEL_GLO_NOISE_L4_NRT_OBSERVATIONS_008_032 describes the noise level of alongtrack measurements delivered. This is a gridded product. One file is provided for the global ocean and those values must be applied for Arctic and Europe products. For Mediterranean and Black seas, one value is given, as described in the following table. SEALEVEL_*_REF20YTO7Y_L4_OBSERVATIONS_008_* is provided temporarily for the needs of users who are still working with the reference period of 7 years instead of 20 years (see V ). This is a gridded product. There are three zones provided: global ocean, Mediterranean and Black seas. For Arctic and Europe areas, the file to be taken into account is the global product. Along-track Sea level anomaly NRT SLA (SEALEVEL_*_SLA_L3_NRT _OBSERVATIONS_008_*) Gridded Sea level anomaly NRT MSLA (SEALEVEL_*_SLA_MAP_L4_NRT _OBSERVATIONS_008_*) Global X X Mediterranean X X Black Sea X X Arctic X Not delivered Europe X Not delivered Table 11: List of the time varying products in NRT c EU Copernicus Marine Service - Public Page 40

42 Gridded Noise on SLA NRT NOISE SLA (SEALEVEL_GLO_NOISE_L4_NRT _OBSERVATIONS_008_*) Gridded Change of reference CHANGE REF SLA (SEALEVEL_*_REF20YTO7Y_L4 _OBSERVATIONS_008_*) Global X X Mediterranean see table 5 X Black Sea see table 5 X Arctic Same as global Same as global Europe Same as global Same as global Table 12: List of the time invariant products in NRT VI.1.1. Delay of the products The availability of the products in near real time is for along-track products: three to twelve hours after the measurement for regional products and 2 hours for global Saral, and Jason-2 products and 3 hours for global Cryosat-2 products. for gridded products: day-0, day-3 and day-6 days. Those products are delivered every day. Three merged maps are produced daily, each with a different delay and quality: A 6-day delay, which represents a final NRT map production, A 3-day delay, which represents an intermediate map production, and a 0-day delay, which represents a preliminary map production, based on IGDR+OGDR production. Then, these maps are replaced when a better quality data is available: At d 0+3, the intermediate map replaces the preliminary map which was produced at d 0. At d 0+3, the final NRT map replaces the intermediate map which was produced at d 0. At d 0+6, the intermediate map replaces the preliminary map which was produced at d 0+3. At d 0+6, the final NRT map replaces the preliminary map which was produced at d 0. c EU Copernicus Marine Service - Public Page 41

43 Figure 10: Three merged maps are produced daily: final map (d-6), intermediate map (d-3) and preliminary map (d0) VI.1.2. Temporal availibility The following table presents the available products by mission and by data period: Near real time products: NRT Jason-2 Cryosat-2 Saral/AltiKa HY-2A Merged Temporal availability Time y/m ongoing y/m ongoing y/m ongoing y/m ongoing NRT-SLA X X X X NRT-MSLA y/m ongoing X VI.2. Delayed Time Products The Delayed Time component of SL-TAC system is responsible for the production of processed HY-2A, Saral/AltiKa, Cryosat-2, Jason-1, Jason-2, T/P, Envisat, GFO, ERS1/2 data in order to provide a homogeneous, inter-calibrated and highly accurate long time series of SLA. DT products are more precise than NRT products. Two reasons explain this quality difference. The first one is the better intrinsic quality of the POE orbit used in the GDR processing. The second reason is that in the DT DUACS processing, the products can be computed optimally with a centred computation time window for OER, LWE. On the contrary in NRT case, "future" data are not available so the computation time window is not centred and therefore not optimal. As for NRT products, improved altimeter corrections and processing algorithms are used: ocean tide model to correct altimeter data, improved methods for orbit error reduction and mapping. As described in V and V time invariant products are also disseminated: c EU Copernicus Marine Service - Public Page 42

44 SEALEVEL_GLO_NOISE_L4_REP_OBSERVATIONS_008_033 describes the noise level of alongtrack measurements delivered. This is a gridded product delivered only on global ocean. For each mission two files are provided: one for filtered products and one for unfiltered products. For Mediterranean and Black seas, one value is given, as described in the following table. SEALEVEL_*_REF20YTO7Y_L4_OBSERVATIONS_008_* is provided temporarily for the needs of users who are still working with the reference period of 7 years instead of 20 years (see V ). This is a gridded product. There are three zones provided: global, Mediterranean and Black seas. Along-track Sea level anomaly REP SLA (SEALEVEL_*_SLA_L3_REP _OBSERVATIONS_008_*) Gridded Sea level anomaly REP MSLA (SEALEVEL_*_SLA_MAP_L4_REP _OBSERVATIONS_008_*) Global filtered X X Global unfiltered X - Mediterranean filtered X X Mediterranean unfiltered X - Black Sea filtered X X Black Sea unfiltered X - Table 13: List of the time varying products in Delayed Time Gridded Noise on SLA REP NOISE SLA (SEALEVEL_GLO_NOISE_L4_REP _OBSERVATIONS_008_*) Gridded Change of reference CHANGE REF SLA (SEALEVEL_*_REF20YTO7Y_L4 _OBSERVATIONS_008_*) Global filtered X X Global unfiltered X Same as filtered Mediterranean filtered see table 10 X Mediterranean unfiltered see table 10 Same as filtered Black Sea filtered see table 10 X Black Sea unfiltered see table 10 Same as filtered Table 14: List of the time invariant products in Delayed Time c EU Copernicus Marine Service - Public Page 43

45 VI.2.1. Delay of the products Daily products are delivered. The availability of the products in delayed time is at the best two months after the date of the measurement. The product generation needs all the GDR data of all the missions to take into account the best corrections as possible. The time delay can be longer in the case of a missing mission. c EU Copernicus Marine Service - Public Page 44

46 VI.2.2. Temporal availibility The following table presents the available products by mission and by data period: Temporal availibility DT-SLA DT-MSLA begin date end date HY-2A 2014/04 y/m X Saral/AltiKa 2013/03 y/m X Cryosat /04 y/m X Jason /10 y/m X Jason-1 geodetic 2012/ /04 X Jason-1 new 2009/ /03 X Jason / /10 X GFO 2000/ /09 X Envisat new 2010/ /08 X Envisat 2002/ /10 X ERS-1/ERS / /04 X Topex new 2002/ /10 X Topex 1992/ /04 X Merged 1992/09 y/m X y/m: those dates are updated regularly (3 to 4 times per year) and are avalaible when you download the data at Jason-1 geodetic orbit: starting 2012/05, Jason-1 new orbit : starting 2009/02, Envisat new orbit : starting 2010/10, T/P new orbit : starting 2002/09, ERS-1: Geodetic phases (E-F) are included. No ERS-1 data between December 23, 1993 and April 10, 1994 (ERS-1 phase D - 2 nd ice phase). The merged products were obtained with the satellites given in figure 11. Moreover, the global attribute in the gridded file called "platform" gives the list of satellites used to compute the map. c EU Copernicus Marine Service - Public Page 45

47 Figure 11: List of satellites in all-sat-merged products c EU Copernicus Marine Service - Public Page 46