WAVE Thematic Assembly Centre:

Transcription

1 WAVE Thematic Assembly Centre: dataset-wav-alti-l3-swh-rt-global-j3 dataset-wav-alti-l3-swh-rt-global-s3a dataset-wav-alti-l3-swh-rt-global-al Contributors: N. Taburet, R. Husson, E. Charles Approval date by the CMEMS product quality coordination team: 23/04/2018

2 CHANGE RECORD When the quality of the products changes, the QuID is updated, and a row is added to this table. The third column specifies which Sections or Sub-sections have been updated. The fourth column should mention the version of the product to which the change applies. Issue Date Description of Change Author Validated By February 2018 All First version of document at CMEMS V4, done from previous SeaLevel-TAC QUID for the wave component R Husson, N Taburet, E Charles Mercator Océan Page 2/ 38

3 TABLE OF CONTENTS I Executive summary... 5 I.1 Products covered by this document... 5 I.2 Summary of the results... 5 I.3 Estimated Accuracy Numbers... 6 II Production system description... 7 II.1 Production center name... 7 II.2 Operational system name... 7 II.3 ABC of the altimeter-derived SWH measurements... 7 II.4 Production centre description for the version covered by this document... 8 II.4.1 Acquisition... 9 II.4.2 Data editing II Editing criteria II.4.3 Calibration II Cross-calibration II Sentinel-3A SAR cross-calibration with Jason II AltiKa cross-calibration with Jason II Absolute calibration II.4.4 Product generation and quality control III Validation framework IV Validation results IV.1 Validation and monitoring of significant wave height IV.1.1 Data availability and spatio-temporal coverage IV.1.2 Data editing monitoring IV.1.3 Multi-mission cross-calibration monitoring IV.2 Additional validation studies to assess processing uncertainties IV.2.1 Assessment of the wave editing performance using the SWH RMS dispersion IV.2.2 Assessment of cross-calibration uncertainties IV Uncertainty associated with the calibration period length IV Uncertainty associated with the crossover time constraint IV Uncertainty associated with the cross-calibration method V System s Noticeable events, outages or changes Page 3/ 38

4 V.1 System version changes V.2 Main constellation events impacting data availability and quality V.3 Historical changes in the calibration and potential impact for users V.3.1 October 2017 System v1.1: new calibration procedure to account for L2 input data versioning.. 34 V.3.2 December 2017 New version of Sentinel-3A L2 production chain (IPF6.10) V.3.3 January 2018 System v1.2: introduction of AltiKa mission and Improvement of S3A / J3 crosscalibration VI Quality changes since previous version VII References Page 4/ 38

5 I EXECUTIVE SUMMARY I.1 Products covered by this document This document describes the quality of the operational (NRT) along-track significant wave height products listed hereafter: Product Area Missions Spatial resolution Temporal resolution Global ocean Jason-3; Sentinel-3A; SARAL/AltiKa Along-track ~7 km (full 1Hz resolution) 10 days to 27 days respectively for Jason-3 and Sentinel-3A (variable with satellite cycle length), drifting orbit (non cyclic) for Altika; products are stored in one file per pass. The number of altimeter data processed by the system varies with time, according to satellites availability. Table 1: summarizes the periods during which the datasets for each mission are available. Temporal availability Begin date End date S3A 2017/07/09 Present J3 2017/07/11 Present AL 2018/01/29 Present Table 1: Temporal period processed by the L3 alti wave chain for the different datasets. Those periods are necessarily shorter than L2 availability presented in Table 4. I.2 Summary of the results The quality of the along-track significant wave height (SWH) product is controlled at each step of the L3 alti wave chain processing. First, editing criteria are applied to the upstream L2 data to select only the best quality data over the ocean surface. 36.7%, 18.5% and 42.8% of measurements are rejected for, respectively, Jason-3, Sentinel-3A SAR mode and AltiKa. Jason-3 is used as the reference mission as it is a conventional altimeter mission, expected to show robust results for SWH measurements. Sentinel-3A SAR mode and AltiKa edited measurements are cross-calibrated with respect to this reference. To determine the cross-calibration relation, crossover points with Jason-3 are computed at global scale on a 138-day period for Sentinel-3A (November 23th 2016 to April 9th 2017) and on a 576-day period for AltiKa (February 17th 2016 to September 14th 2017). Table 2 summarizes the performance of the cross-calibration in terms of bias and standard deviation of the differences between the reference mission and the secondary missions. Page 5/ 38

6 Cross-calibrated Before cross-calibration After cross-calibration missions Bias Standard dev. Bias Standard dev. Sentinel-3A / Jason cm 23.5 cm 0.5 cm 21.8 cm AltiKa / Jason cm 21.1 cm cm 20.8 cm Table 2: Bias and standard deviation between Jason-3 and secondary missions SWH, before and after the cross-calibration step. Absolute calibration with respect to buoys is finally applied to all SWH cross-calibrated data. I.3 Estimated Accuracy Numbers The noise measurement error (i.e. uncorrelated error) is given in Table 3. The average noise is the uncertainty associated with the 1-Hz significant wave height estimates. It is the standard deviation of the high frequency (HF) measurements used to compute the 1 Hz value, divided by the square root of the number of HF points. AltiKa mission presents a smaller noise as a result of its high-frequency sampling of 40 Hz (20 Hz forjason-3 and Sentinel-3A). Mission Average noise (cm) Jason-3 12 Sentinel-3A (SAR mode) 9 AltiKa 6 Table 3: Mean 1Hz noise measurement observed for the different altimeters on along-track (L3) WAVE- TAC products. Unit: cm rms. Page 6/ 38

7 II PRODUCTION SYSTEM DESCRIPTION II.1 Production center name WAVE-CLS-TOULOUSE-FR II.2 Operational system name The operational chain is given a generic name: the L3 alti wave chain. II.3 ABC of the altimeter-derived SWH measurements The altimeter sends a spherical radar signal in the direction of the nadir. This signal is reflected by the sea surface and goes back to the satellite. The analysis of the returned signal allows the calculation of the time needed by the signal to go and come back, i.e. the distance satellite-sea surface. The sea state surface elevation distribution impacts the speed at which the return signal is fully returned to the satellite. Hence, the Significant Wave Height (SWH) over ocean surfaces is determined from the slope of the front in the radar altimeter wave form. The higher the waves, the more the returned signal is spread in time. Hence, a long delay between the first returns and a full signal return will result in a long shadow in the wave form, which then indicates a high sea state (Figure 1 and Figure 2). The term Significant Wave Height (SWH or H s ) refers to the mean wave height of the highest third of the waves (also sometimes denoted H 1/3 ). Figure 1: Formation of an echo over a sea surface with waves for conventional altimetry Page 7/ 38

8 Figure 2: The altimeter waveform II.4 Production centre description for the version covered by this document The system s primary objective is to provide operational products of calibrated significant wave height (SWH) data for Jason-3, AltiKa and Sentinel-3A missions. The processing sequence can be divided into 4 main steps, illustrated in Figure 3 and described in the next Sub-sections: Acquisition; Data editing; Calibration; and Product generation. Page 8/ 38

9 L2-ACQUISITION CRON Start L3 production DataUpdate RPC Client L3-PRODUCTION DAD SAD PRE-PROCESSING (data acquisition) DATA UPDATE Internal database EDITING (valid mesures selection) L2 Files CALIBRATION FILES GENERATION (NetCDF ) PRODUCT QUALITY MONITORING L3 Product Quality Report L3 NRT (near real time) II.4.1 Acquisition Figure 3: L3 Alti wave production component The altimeter measurements used in the system consist in Near-Real-Time (OGDR or NRT) Level 2 products from different missions. Their source, delay and period of availability are summarised in Table 4. Mission characteristics are presented in Table 5. We point out that L2 availability period is longer than L3-wave products produced since their introduction in the L3 WAVE-TAC system (see Table 1). Mission Type of Availability Source product delay Period of availability Jason-3 OGDR EUMETSAT/NOAA ~3h 2016/12/13 (cycle 12) - present Sentinel-3A NRT ESA/EUMETSAT ~3h 2016/02/17 (cycle 1) - present AltiKa OGDR EUMETSAT ~3h 2013/03/14 (cycle 1) - present Table 4: Source, delay and period of availability of the different altimeter data Page 9/ 38

10 Mission Cycle duration (days) Latitude range ( N/S) Number of tracks per cycle Inter-track distance at equator (km) Sunsynchronous Technology Jason-3 10 ± ~315 No LRM Sentinel-3A 27 ± ~100 Yes SAR + PLRM AltiKa Non-cyclic (since July 2016) ± (per pseudo cycle) ~75 Yes LRM Table 5: Altimeter mission characteristics The acquisition processing has two main functions: acquisition and synchronization of dataflow as illustrated in Figure 4. File acquisition The purpose of the acquisition is to acquire new L2 files and new ancillary data (AUX files) needed to compute the products (orbit file, external corrections, etc.) for each data source. Each L2 file acquired can be updated with its availability date read on the source server. This ensures using the most recent input files and avoiding unnecessary updates. Data synchronization The synchronization function is synchronizing L2 data with all ancillary data (AUX files) needed to process L3 data. Once the L2 data and all the associated ancillary data are available, they can be used for L3 production. L2 NRT (near real time) AUX (ancillary data) L2-ACQUISITION ACQUISITION DATA SYNCHRONIZATION (ADEA) notification Files L3-PRODUCTION Figure 4: L2 acquisition processing Page 10/ 38

11 II.4.2 Data editing II Editing criteria Quality Control on the input L2 data is a critical process applied to guarantee that the system uses only the most reliable altimeter data. The system is supplied with L2 products that contain data directly derived from altimeter measurements (e.g. range, sigma0, etc.) as well as geophysical data (e.g. dry tropospheric correction, significant wave height, etc) and flags (e.g. surface type, ice presence, etc.). These values are provided at high (20 Hz for Jason-3) and low (1 Hz) frequency. Only the 1 Hz data are used in the L3 alti wave system. Data are selected as valid or invalid using a combination of various criteria such as quality flags and parameter thresholds (see Table 6 for details). These criteria are adapted from the ones used for the Sea Level Anomaly (e.g. Aviso/SALP 2016). Only criteria related to retracking derived values were selected. Geophysical parameters (e.g. tropospheric corrections) do not intervene in the SWH estimation. For Sentinel-3A the criteria on the off-nadir angle is not activated since this value is not derived from the retracking in SAR mode and therefore its value does not provide information about data quality. Parameter Method Jason-3 Sentinel-3A SAR AltiKa Ice Flag Flag value Valid value: 0 or 5 Valid value: 0 Valid value: 0 Surface type Flag Flag value Valid value: 0 or 1 Valid value: 0 Valid value: 0 Swh [m] Threshold Min: 0 max: 30 Min: 0 Max: 30 Min: 0 Max: 30 Sigma0 [db] Threshold Min: 9.38 max: Min: 5 Max: 28 Min: 3 Max: 30 Square off-nadir angle Threshold Min: -0.2 max: 0.64 N/A Min: -0.2 Max: Wind speed [m/s] Threshold Min: 0 Max: 30 Min: 0 Max: 30 Min: 0 Max: 30 Orbit - range [m] Threshold Min: -130 Max: 100 Min: -130 Max: 100 Min: -130 Max: 100 Sigma0 standard deviation [db] Threshold Min: 0 Max: if distance to shoreline <50 km: 2.5 else: 1 Threshold Min: 0 Max: *swh+0.2 Min: 0 Max: 0.7 Min: 0 Max: 1 Range standard deviation [m] Min: 0 Max: 0.02*swh+0.12 Min: 0 Max: 0.2 swh_numval_ku Threshold Min: 10 Min: 18 Min: 20 swh_rms_ku Threshold Min: 0 Min: 0 Min: 0 Table 6: Flag and threshold editing criteria for the different missions These editing criteria were applied to data acquired from November 23th, 2016 to April 9th, 2017 (about 140 days). The resulting percentages of rejected measurements are provided in Table 7. The very low rejection level due to the surface type flag on S3-A SAR data can be explained by the use of the ocean-only Sentinel-3A products. S3A land and ocean products present a small overlap, of the order of 300 km, explaining the 2% rejection level due to the surface type. Edited measurement percentage is higher for AltiKa and can be explained by its coverage at higher latitudes (81.5 N/S compared to 66 N/S for Jason-3), inducing a high percentage of ice flags. Thanks to the high quality of current missions, the threshold criteria reject a small percentage (less than 4%) of altimeter Page 11/ 38

12 measurements for all missions. The percentage of threshold edited measurements is expected to present an intra-annual variability of a few tenths of a percent, as seen in CalVal studies. Parameters Jason-3 Sentinel-3A SAR AltiKa Ice Flag Surface type Flag (*) 34 Combined Flags Swh Sigma Square off-nadir angle 0.56 N/A 0.29 Wind speed Orbit - range Sigma0 standard deviation Range standard deviation Combined thresholds All criteria Table 7: Percentage of rejected measurements estimated over J-3, S3-A and AL data acquired from November 23th 2016 to April 9th 2017 (* see text for explanation). We compared the statistics of rejected measurements at Jason-3 and Sentinel-3A 3-hour cross-overs using the wave editing and the sea level editing approaches. Results are similar in terms of significant wave height dispersion between the two missions but allows having about 6% more points due to the relaxed constraints on geophysical parameters in the wave editing approach. II.4.3 Calibration Calibration is divided in two main steps (see Figure 5): cross-calibration on reference mission and absolute calibration on in-situ data. The first step consists in homogenising the data from the different missions. Significant wave height measurements of every single mission are calibrated on those of a reference mission (Jason-3). The second step consists in applying a correction computed between the reference mission and in-situ measurements provided by buoys. Finally, another calibration step can be added in the process when L2 upstream products evolve for a mission already implemented in the L3 alti wave chain (see Figure 5, L2 version upgrade in yellow). A new calibration for the physical variables of interest is determined between the current and the upcoming L2 version and is added to the existing calibration of this mission in the L3 alti wave chain. The next Sub-sections describe the computation of the two main calibrations: cross-calibration and absolute calibration. An history of the L3 alti wave chain calibration changes is given in Section V.3. Page 12/ 38

13 II Cross-calibration Figure 5: Description of the calibration process Cross-calibration consists in determining the relation between the significant wave height measurements provided by two different missions. This relation is determined on a representative number of collocated measurements and then used in the operational system to homogenise the missions with respect to the reference one. Such a relation is expected to remain valid as long as instrumental drifts are not detected or ground segment evolutions do not affect the L2 products in input of the operational system. Should one of these evolve, another cross-calibration relation should be computed and implemented into the operational system. Jason-3 is used as the reference mission as it is a conventional altimeter mission, expected to show robust results for SWH measurements. Two different methods of collocation can be considered, depending on the orbit of the mission to be calibrated with respect to the orbit of the reference mission. The first one is applied during the tandem phase, if it exists, between two missions: both satellites are on the same orbit separated by a few minutes. A very large number of spatially collocated measurements are therefore available for cross-calibration. The second method is employed when the two missions are on different orbits or no validation phase is available. Crossover points between the two orbits are determined. For SWH measurements calibration, only crossover points with a time difference lower than 3 hours are considered. This short delay ensures that both missions observe a scene that did not significantly evolve (when a longer dataset archive is available, this time difference can be lowered to 1 hour). The 1 Hz along track data for each mission is then interpolated at the selected crossover points. The interpolation technique consists in spline approximation and accounts for the average noise associated with SWH measurements. Such values are given in Table 3 and correspond to the uncertainty on the 1 Hz significant wave height values computed from the high frequency values (20 Hz for Jason-3 and Sentinel-3A and 40 Hz for AltiKa). Once the two missions measurements are collocated, the differences between the reference mission and the secondary mission significant wave heights are computed. The bias is plotted as a function of the secondary mission wave height in order to provide a height-dependent bias correction. The next step consists in fitting a polynomial function to the distribution of this bias. This function is finally Page 13/ 38

14 inserted in the L3 alti wave chain to be systematically applied to all L2 edited measurements of the secondary mission. The following Sub-sections present the computation of the cross-calibration between the secondary missions Sentinel-3A and AltiKa and the reference mission Jason-3. Some tests on cross-calibration uncertainties are presented in Section IV.2.2. II Sentinel-3A SAR cross-calibration with Jason-3 Sentinel-3A crossover points with Jason-3 were computed on a 138-day period (November 23th, 2016 to April 9th, 2017). The starting date corresponds to the beginning of the production of Sentinel-3A product with the Samosa 2.3 ocean retracking used to determine the wave height from the waveform. Figure 6 presents the spatial distribution of the valid crossover points after editing. The number of points is larger at high latitudes in the southern hemisphere, allowing sampling higher waves associated with extra-tropical storms. Figure 6: Spatial distribution of Sentinel-3A and Jason-3 crossover points. Only valid points after editing are displayed. Top: histogram of the selected points. Right: Number of points and mean SWH valid values as a function of latitude. The representativeness of the crossover points with respect to the ensemble of valid along-track points is checked by comparing the Jason-3 SWH distribution for the two ensembles of points (Figure 7). The distribution at the crossover points is skewed towards larger SWH values due to the larger density of crossover points at high latitudes where the mean SWH is larger than in the inter-tropical band. Despite these distribution differences, crossover points sample all range of significant wave height values from 0.5 to 6 m. Outside this interval, the population in each bin of 10-cm width is smaller than 10 points and the cross-calibration fit is likely to be less reliable. Page 14/ 38

15 Figure 7: Valid Jason-3 points distribution over the cross-calibration period with S3-A. This cross-calibration differs from the others since the wave height bias between S3A SAR and J3 presents a relation that cannot be correctly fitted by a linear function. This is a known result when using the S3A SAR data generated by the L2 PDGS chain. The wave height bias between the two modes exhibits a higher order dependency on wave height and is better fitted by a second-order polynomial fitting function (Figure 8). About 5000 collocated measurements are available, however, a second-order polynomial function requires more collocated points than a linear function, especially at the tail of the distribution. We tested the usual direct cross-calibration method (Figure 8) over two wave height ranges: [0-6 m] (orange) and [0-12 m] (green). Both polynomial functions are in good agreement over the [0-6 m] range but significantly differ for larger values. Residuals are presented in the bottom plot. As expected, considering the whole population, the polynomial function computed over the [0-6 m] range (orange) poorly samples high wave heights and present a large bias and dispersion. The mean bias per bin when considering the [0-12 m] fit still presents a mean bias larger than 6 cm (green). Page 15/ 38

16 Figure 8 : Top: Blue dots are the median of the difference between S3-A SAR and J3 SWH values per 10 cm bin. Error bars represent the standard deviation of the difference inside each bin. The orange and green curves represent second order fitting polynomials, respectively over the [0-6 m] and [0-12 m] ranges. Bottom: Residuals between the median and the fits. Green dots are the residuals between the [0-12 m] direct fitting polynomial and the whole population, while orange are the residuals between the polynomial adjustment determined on [0-6 m] and the whole population. Therefore, we take advantage of the simultaneous SAR and PLRM measurement modes of S3-A to apply a two-stage cross-calibration method: Computation of the linear fit between S3-A PLRM and J3 measured wave heights (Figure 9) Computation of the second-order polynomial fit between S3-A SAR and PLRM measured wave heights over a large number of collocated measurements (Figure 10) The polynomial and linear functions are applied consecutively to S3-A SAR measured wave heights to intercalibrate S3-A SAR measurements over the reference mission J3 wave height distribution (Figure 11). First, we compute the linear fit between S3-A PLRM and J3 measured wave heights at crossover points (Figure 9). Although the dispersion remains important due to the poor statistics at large SWH values, the mean bias is of the order of +/-3 cm. Both fits on the [0-6 m] or [0-12 m] ranges (orange and green curves respectively) are in good agreement and less affected for high waves than with a direct fit of the SAR over the Jason-3 values. We select the linear correction function computed over the [0-6 m] range, where most of the wave population lies (see V.3 for more details): Corr(S3A PLRM /J3) = H Equation 1: Linear correction for cross-calibration of S3A PLRM and J3 LRM significant wave height Page 16/ 38

17 Figure 9: Top: Median of the difference between S3-A PLRM and J3 SWH values per 10 cm bin. Error bars represent the standard deviation of the difference inside each bin. The orange and green curves represent respectively the linear fits over the [0-6 m] and [0-12 m] ranges. Bottom: Residuals between the median and the fits. The mean bias is of the order of 3 cm. Secondly, using the simultaneously valid PLRM and SAR S3-A measurements over the same period (about 7 million points), the SAR vs PLRM bias fitting function can therefore be accurately computed for wave heights larger than 6 m (Figure 10). Regarding the wave heights larger than 12 m, not enough measurements are available to correctly compute the fitting function. We also discard wave heights smaller than 1 m, as they present a behaviour different from the rest of the distribution (see V.3 for more details). Therefore, the cross-calibration correction for S3A SAR vs PLRM is computed over the [1-12 m] range and is assumed to remain constant and is extrapolated for values outside this range: Corr(S3A SAR /S3A PLRM ) = H H Equation 2: 2 nd -order correction for cross-calibration of S3A SAR and PLRM significant wave height Page 17/ 38

18 Figure 10: Top: Median of the difference between S3-A SAR and PLRM SWH values per 10 cm bin. Error bars represent the standard deviation of the difference inside each bin. The orange curve represents the second order fitting polynomial. Bottom: Residuals between the median and the fit. The mean over the [1-12 m] range shows that the residuals are unbiased. The dispersion is of the order of 3 cm. Finally, combining the S3-A SAR vs PLRM and S3-A PLRM vs J3 bias corrections, a global fitting function allows to cross-calibrate S3-A SAR on J3 SWH. Figure 11 shows that this two-stage cross-calibration method (red curve) is in good agreement with the direct cross-calibration method (green curve) on the [0-6 m] range. The direct method is nevertheless much more uncertain on the [6-12 m] range due to a much smaller number of measurements within that range. For both methods, uncertainties are driven by the fit between J3 and S3A at crossovers, due to a limited number of crossover points. The residuals bias is of the order of -3 cm for the two-stage method, smaller than the 10-cm bias obtained with the direct method. The two-stage cross-calibration process is therefore implemented in our system to calibrate S3-A SAR on J3. Page 18/ 38

19 Figure 11 : Top: Median of the difference between the S3-A SAR SWH value and the J3 values per 10 cm bin. Error bars represent the standard deviation of the difference inside each bin. The red curve represents the fitting function obtained with the two-stage cross-calibration method, and the green curve is the fitting function obtained with the direct method. Bottom: Residuals between the median and the two-stage fitting function (red) as well as with the direct method (green). II AltiKa cross-calibration with Jason-3 Cross-calibration of AltiKa significant wave heights with Jason-3 was performed using the crossover point method with a time difference lower than 3 hours. The interpolation at crossovers was performed, accounting for a 6-cm noise associated with the 1-Hz SWH measurements 1. AltiKa / Jason-3 values at their 3-hour crossover points were computed over a 576-day period (February 17 th, 2016 to September 14 th, 2017). The spatial distribution of such valid points is presented on Figure 12. Similarly to S3A/J3, the number of crossover points is larger at high latitudes therefore sampling a larger SWH population. As a result, the distribution of the SWH at crossover points is slightly skewed towards larger values when compared to the one computed along track (Figure 13). Crossover point distribution nevertheless samples all ranges of SWH values from 0.5 to 6 m. 1 Smaller than for Jason-3 and Sentinel-3A due to the 40-Hz high frequency sampling of AltiKa. Page 19/ 38

20 Figure 12: Spatial distribution of AltiKa and Jason-3 crossover points. Only valid points after editing are displayed. Top: histogram of the Altika SWH values at collocated points. Right: Number of points and mean SWH valid values as a function of latitude. Figure 13: Valid AltiKa SWH values distribution over the cross-calibration period The blue dots in Figure 14 (top) represent the median of the difference between AltiKa and Jason-3 SWH values inside each 10-cm bin of the Altika population. A linear fit is performed to determine the bias as a function of the significant wave height. The selected fitting function is the one computed over the [0-6 m] range (orange) as it samples most of the population and presents a low bias (5 mm) and a similar dispersion to the linear fit over the [0-12 m] range (around 20 cm): Page 20/ 38

21 Corr(AL/J3) = H Equation 3: Linear correction for cross-calibration of AL and J3 significant wave height Figure 14: Top: Median of the difference between AltiKa and J3 SWH values at crossover points per 10- cm bin over the period February 17th September 14th Error bars represent the standard deviation of the difference inside each bin. The orange and green curves represent linear fits over different SWH ranges. Bottom: Residuals between the median and the fits. II Absolute calibration Once inter-mission biases are removed, using the cross-calibration corrections described in the previous section, an absolute calibration correction is applied to all missions. This absolute calibration aims at correcting the biases between in-situ measurements and satellite altimetry. All the missions are cross-calibrated on the reference mission Jason-3. Therefore, the absolute calibration is computed from the comparison of Jason-3 significant wave heights to buoy measurements at collocated points. According to results from Queffeulou [2016], performances for Jason-3 are very similar to those given by Jason-2 and can thus be applied to compensate for systematic errors. The linear correction is given below [Queffeulou and Croizé-Fillon 2017]: H corr (J2/buoys) = 149 H Equation 4: Linear correction for absolute calibration of Jason-2 SWH with respect to buoy measurements Comparisons between Jason-3 and Jason-2 along-track 1 Hz collocated measurements during Jason-3 commissioning (same track, 80s difference between the two altimeters), were performed to compare sea state sensed by the two altimeters at the same geographical location (Figure 15). The left plot Page 21/ 38

22 shows that Jason-3 and Jason-2 1-Hz collocated SWH are in very good agreement. The regression line is very close to the unity. The bias is less than 2 mm and the RMSE is about 19 cm. The right plot shows a symmetrical distribution of the SWH which indicates similar precisions. Figure 15: JASON-3 and JASON-2 1-Hz collocated SWH (left) and SWH RMS (right) (SWH RMS filtering applied). Extracted from Queffeulou [2016]. II.4.4 Product generation and quality control The bias correction described in previous section is applied to SWH values issued from all missions. One NetCDF file per pass is generated for each mission. As explained in Section II.4.1, due to the successive treatments, as soon as elementary data are available, new L3 products are generated and contain the most recent and complete set of data. The L3 along-track products contain the fields described in Table 8. Standard name Long name NetCDF Type Units time time (sec. since ) TAI seconds since double (International Atomic Time) :00:00.0 latitude latitude int 10-6 deg longitude longitude int 10-6 deg sea_surface_wave_significant_height significant wave height on main altimeter frequency band short 10-3 m Table 8: Along-track significant wave height variable and dimensions included in each L3 NetCDF file Daily automated controls are performed, and upon generation, warnings are sent to operators. Quality control reports are also generated once a day and analyzed by altimetry experts (internal validation, those reports are not disseminated). Sections III and IV.1 present the diagnostics implemented in these reports. Page 22/ 38

23 III VALIDATION FRAMEWORK The validation aims to control the quality of the produced significant wave height and the performances of the key processing steps. Different points are assessed by the validation task: The data availability and spatial/temporal coverage The data editing monitoring The multi-mission cross-calibration monitoring The ocean signal consistency Table 9 lists the different metrics that are used. They mainly consist in an analysis of the SWH field at different steps of the processing and in checking the consistency of SWH along the tracks of different altimeters. Uncertainties affecting the produced significant wave height can also be investigated through specific studies carried out at different steps of the processing. Their results are presented in Section IV.2. Page 23/ 38

24 Name Description Ocean parameter Supporting reference dataset Quantity L3 SWH_L2-NC-AVAIL-<period> SWH_L2-NC-VALID-<period> Number of altimeter measurements missing/available Number of altimeter measurements valid/invalid Significant Wave Height Significant Wave Height None None Missing data are identified over the data flow processed Temporal evolution on the number of measurements on a daily/weekly/monthly basis and/or along each track of the altimeter considered. As for Saral/AltiKa the current method is not suitable for drifting mission and is therefore not implemented. Method is planned to be updated to track AltiKa missing data. Valid/rejected data are identified over the data flow processed Temporal evolution on the number of measurements on a daily/weekly/monthly basis and/or along each track of the altimeter considered. SWH_L2-NC-MEAN_T SWH signal monitoring Significant Wave Height None Temporal evolution on the weekly-averaged significant wave height estimated between +/- 66 latitude estimated over several months for each mission for all/valid data. The associated temporal evolution of the number of all/valid samples should also be attached for mission with high latitude sampling. SWH_L2-NC-ALT-MEAN_T-XOVER SWH_L2-NC-ALT-CRMSD_T-XOVER POS_SWH_L2-CLASS3-ALT-VALID-XOVER- <period> SWH-M-NC-MEAN-GLB SWH-M-NC-STD-GLB SWH-M-NC-VALID-GLB SWH-M-NC-REJ-GLB SWH differences at mono- and multimissions crossover positions SWH signal monitoring Significant Wave Height Significant Wave Height None None Temporal evolution of the weekly-averaged mean difference between two SWH measurements corresponding to altimeter tracks cross-over positions (typically estimated over several months). The performances of the product before and after calibration (inter-calibration, absolute calibration wrt. in situ) are compared. Temporal evolution of the weekly-averaged standard deviation of the difference between two SWH measurements corresponding to altimeter tracks cross-over positions (typically estimated over several months). The performances of the product before and after calibration (inter-calibration, absolute calibration wrt. in situ) are compared. Temporal evolution of the weekly-averaged of the number of SWH measurements corresponding to altimeter tracks cross-over positions (typically estimated over several months). Global map of the averaged along-track SWH (L3) over a month (2x2 grid, for each cell). Global map of the standard deviation of along-track SWH (L3) over a month (2x2 grid) Global map of number of along-track SWH valid samples (L3) over a month (2x2 grid) Global map of number of along-track SWH rejected samples (L3) over a month (2x2 grid) Table 9: List of the metrics used for WAVE-TAC products operational validation Page 24/ 38

25 IV VALIDATION RESULTS Validation metrics are used operationally to monitor the quality of the produced significant wave height. These metrics are listed in Table 9 and described in detail in the CMEMS-QP-WAVE-ScVP document. In Section IV.1, examples of their application over different time periods are presented. Assessment of the uncertainties affecting WAVE-TAC products are also completed by specific studies carried out at different steps of the processing, presented in Section IV.2. IV.1 Validation and monitoring of significant wave height IV.1.1 Data availability and spatio-temporal coverage Figure 16 presents the application of the data availability diagnosis over the period used for crosscalibration between S3A and J3. The left plot represents the available Level-2 data over the whole period whereas the right plot represents the missing data. Information about the last day of the period is highlighted on both plots in red. Such a diagnosis on the last day is of great interest in the offline validation process that runs every day. Figure 16: L2 data availability for Jason-3 over the calibration period. Left: available data. Right: missing data. Blue = whole period, Red = last day IV.1.2 Data editing monitoring Figure 17 presents the application of the data editing diagnosis over the whole period used for intercalibration between S3A and J3 and also provides information about the last day of the period (in red). Figure 18 presents the application of the data editing diagnosis for ice flag and thresholds criteria indicated in Table 6. Page 25/ 38

26 Figure 17: Position of the invalid (valid) measurements. Red stands for the last processed day. Figure 18: Position of the measurements invalidated on (left) the ice flag criterion and (right) the standard deviation of the range criterion. IV.1.3 Multi-mission cross-calibration monitoring A qualification test of AltiKa cross-calibration with Jason-3 is performed over a 23-day period (November 16 th, 2017 to December 8 th, 2017) in order to assess the quality of the calibration over a different period from the one used to establish the cross-calibration. It consists in computing, for each day, the previous 7-day averaged differences between the calibrated Altika and Jason-3 SWH at crossover for collocation time intervals of 3 hours. As presented in Figure 19 the differences at crossover after calibration (green curve) are centred around 0 with daily variations of the order of 1-2 cm. Page 26/ 38

27 Figure 19 : Mean difference at crossover for a 3-hour time constraint. Blue curve represents the difference before calibration (as in the L2 products). The green curve represents the difference after calibration (as in the L3 wave products) IV.2 Additional validation studies to assess processing uncertainties IV.2.1 Assessment of the wave editing performance using the SWH RMS dispersion Queffeulou [2016] showed that the quality of significant wave height measurements can be given by the shape of the joint distribution of the 1 Hz SWH value (mean value over 1 s, calculated from high resolution 20 Hz products) and the associated standard deviation (SWH RMS). They proposed a criterion based on a maximum SWH RMS as a function of SWH to eliminate potentially erroneous data. Figure 20 illustrates the dispersion of Jason-3, Sentinel-3A and AltiKa SWH RMS before (left) and after (right) editing. It shows that the wave editing significantly decreases the quantity of data in the high SWH RMS area and therefore increases the quality of the retained SWH data. Figure 20 also exhibits a non-linear behaviour for SWH < 1.5 m, unchanged by the wave editing. This feature was also observed by Queffeulou [2016] for all altimeters and may be due to waveform processing. Page 27/ 38

28 (a) (b) (c) (d) (e) (f) Figure 20: Joint distribution of the 1 Hz SWH value (mean value over 1 s, calculated from high resolution 20 Hz products) and the associated standard deviation (SWH RMS) before (left) and after (right) editing, for (a, b) Jason-3; (c, d) Sentinel-3A SAR; (e, f) AltiKa. Page 28/ 38

29 IV.2.2 Assessment of cross-calibration uncertainties IV Uncertainty associated with the calibration period length The same AL/J3 cross-calibration process as in Section II was performed over a 138-day time period (November 23 th 2016 to April 9 th 2017) to obtain an estimate of the uncertainty on the calibration associated with the time period selection (Figure 21). The residual dispersion when considering the fit over the [0-6 m] range increases from 4.1 to 5.5 cm. When comparing both [0-6 m] fits over the two periods, the difference ranges from 0.6 cm for SWH=0 m to 3.2 cm for SWH=12 m, providing an estimate of the uncertainty associated with the calibration period length of the order of a few cm for largest wave heights. Figure 21: Same as Figure 14 but with a shorter time period: November 23th 2016 to April 9th 2017 IV Uncertainty associated with the crossover time constraint In Section II , the cross-over calibration between AltiKa and Jason-3 was computed over crossover points collocated with a time constraint of 3 hours. The dependency with crossover time constraint was eventually assessed by imposing a maximum time difference of 1 hour at crossover. As presented in Figure 22 and Figure 23, this decreases the number of crossover points by a factor of about 3 with respect to a 3-hour time constraint. When comparing the [0-6 m] linear fits using the 1- hour (Figure 24) and 3-hour (Figure 14) time constraints, the difference ranges from 0.4 cm for SWH=0 m to 6 cm for SWH=12 m, providing an estimate of the uncertainty associated with the crossover time constraint of the order of the cm for largest wave heights. Page 29/ 38

30 Figure 22: Spatial distribution of AltiKa and Jason-3 crossover points. Only valid points after editing are displayed. Top: histogram of the Altika SWH values of the selected points. Right: Number of points and mean SWH valid values as a function of latitude. February 17th 2016 to September 14th 2017, 1-hour crossover constraint. Figure 23: Valid AltiKa SWH distribution over the cross-calibration period (February 17th 2016 to September 14th 2017). 1-hour crossover constraint. Page 30/ 38

31 Figure 24: Same as Figure 14 but with a 1-hour crossover time constraint IV Uncertainty associated with the cross-calibration method Other studies carried out a cross-calibration between Jason-3 and AltiKa. We compare here the results obtained by Queffeulou [2016] with the cross-calibration function used for this study. Queffeulou [2016] applied some editing criteria based on SWH RMS dispersion and compared AltiKa and Jason-3 significant wave heights over 1-hour collocated cells (Figure 25). Expressing J3 as a function of AL, they obtained the following calibration correction H corr = H The one obtained in the present study with 1-hour crossover points is H corr = H (Figure 26). Their difference spans from 1.5 cm to -3.7 cm over the [0-12m] range, with 0.6 and -1.1 cm difference at, respectively, SWH=2 m and 6 m (0.30% and -0.19% uncertainties). This confirms the excellent agreement between the two studies despite their different periods and point selection criteria. The bias (defined as the mean of the difference) between the AL and J3 values is 5.4 cm and the dispersion is 15.6 cm (Figure 26). This agrees with Queffeulou s results (bias J3/AL = -5.8 cm and dispersion 22.7 cm). In the present study the dispersion is smaller due to the comparison of the SWH at the crossover position while Queffeulou [2016] values result from collocation within a 1-hour time window, not restricted at the exact crossover position. Adding a 50-km maximum distance criterion they mentioned the dispersion reduces to cm in agreement with our 15.6 cm dispersion value. Their slightly smaller value may be due to a more restrictive editing that includes more restrictive criteria on the standard dispersion of the high frequency wave height estimates used to compute the 1Hz significant wave height value. When using the formula derived from Figure 24 (fit of the difference AL-J3 at 1-hour crossover points); the differences with the Queffeulou [2016] results spans from -5.1 cm to cm over the [0-12 m] interval, with a 2.1 and -5.1 cm difference at SWH= 2 and 6 m (5% and 0.85% uncertainties). This different fitting method also provides results in good agreement with Queffeulou [2016]. This comparison provides us with an estimate of the uncertainty arising from the fitting method (scatter plot VS differences per bins), that is of the order of 5-10 cm. The calibration uncertainty is therefore dominated by the fit method rather than the data selection period length (providing this one is still long enough) or the 1- or 3-hour constraint at crossover. Page 31/ 38

32 Figure 25: SARAL SWH comparison with Jason-3 for 1 Hz, 1-hour collocated data. From Queffeulou [2016] Figure 26: SARAL SWH comparison with Jason-3 valid 1 Hz measurements at less than 1-hour crossover points. Mean difference between the AL and J3 values is 5.36 cm with a 15.6 cm dispersion. Page 32/ 38

33 V SYSTEM S NOTICEABLE EVENTS, OUTAGES OR CHANGES This section is dedicated to track and describe changes that may occur in the current operational system: due to outages, version or upstream data changes (e.g. addition or loss of a satellite). V.1 System version changes System version Date of Entry in Service of the change V 11/07/2017 V1.1 02/10/2017 V1.2 29/01/2018 Description of the change Entry into Service of the L3 alti wave chain with Jason-3 and Sentinel-3A missions New calibration procedure to account for L2 input data versioning (anticipation of future L2 version changes) Integration of SARAL/AltiKa mission Improvement of S3A / J3 cross-calibration correction Table 10: L3 alti wave chain version changes. Impact on product quality? Yes No Yes V.2 Main constellation events impacting data availability and quality Different events can lead to a change in data availability and quality. Such events are usually: A change in the altimeter constellation: the loss or introduction of an altimeter in the constellation directly impacts the number of altimeter measurements available. For a specific platform, a reduction of the number of altimeter measurements available in input of the L3 alti wave system processing. This can be linked with an anomaly onboard the platform or on the ground segment, preventing the data reception and impacting the L0 to L2 processing. It can also be induced by an abnormal acquisition by the L3 alti wave system. An increase of invalid measurements in input of the L3 alti wave system processing. This is usually linked with specific platform events (e.g. maneuvers) but can also be induced by L0-L2 processing anomalies or specificities. In some rare cases, abnormal acquisition by the L3 alti wave system can also lead to an abnormal data selection. A change in the L0 - L2 processing can lead to changes in the quality of the measurements in input of the L3 alti wave system processing. For example, new versions of L2 upstream products are regularly released, to account for state-of-the-art corrections and developments of this upstream processing. Currently, 3 altimeters constitute the altimeter constellation available in NRT. Jason-3 (J3) is the reference mission and the oldest in the constellation. Page 33/ 38

34 Sentinel-3A (S3A). SARAL-DP/AltiKa (AL). Table 11 summarizes the main events affecting the data availability in NRT conditions. Date Platform Event 11/07/2017 J3, S3A Introduction of Jason-3 and Sentinel-3A missions in the L3 wave alti chain 13/12/2017 S3A New version of the Sentinel-3A L2 production chain (IPF 6.10) 29/01/2018 AL Introduction of SARAL/AltiKa mission in the L3 wave alti chain Table 11: Main events affecting the data availability in NRT conditions V.3 Historical changes in the calibration and potential impact for users V.3.1 October 2017 System v1.1: new calibration procedure to account for L2 input data versioning As evolutions can be brought to the upstream L2 production chain, a different cross-calibration correction may be required to ensure consistent performances of L3 products. The computation of this correction requires an early and simultaneous access to the new and previous versions of L2 data in order to determine a new cross-calibration correction between the two L2 versions. This correction is added to the existing correction computed from the cross-calibration with the reference mission. In order to simplify the change or addition of new calibrations, the L3 alti wave chain was modified to use an abaque file, providing the calibration to be applied as a function of both SWH and L2 file version The resolution of the abaque SWH bins is 5 cm. This version v1.1 of the L3 alti wave chain was implemented in October 2017, in order to anticipate a new version of the Sentinel-3A L2 production chain (detailed in paragraph V.3.2). The implementation of this new version has no impact on the production. V.3.2 December 2017 New version of Sentinel-3A L2 production chain (IPF6.10) A new version of the Sentinel-3A L2 production chain (IPF 6.10) is operated at EUMETSAT since December 13 th, 2017 in replacement of the IPF 6.07 version. The 6.10 version presents a modification in the SAMOSA retracking with respect to v6.07, impacting the SAR significant wave height. Consequently, the calibration correction embedded in the L3 alti wave chain was updated to account for this IPF version change. Valid SWH values from both IPF versions were compared during a 20-day period. As all points from IPF v6.10 and IPF v6.07 are collocated, a 20-day period is long enough to perform the cross-calibration. The significant wave height bias between the two L2 versions is presented on Figure 27. A secondorder polynomial function is fitted to this bias and added to the existing Sentinel-3A calibration correction abaque. Page 34/ 38

35 Figure 27: Differences between the IPF versions v6.10 and v6.07 of L2 S3A SAR significant wave height before polynomial adjustment This joint upgrade of L2 IPF version and L3 alti wave chain (from v to v1.1) has little impact on the produced Sentinel-3A L3 data. We plotted the differences between L3 SWH produced with IPF v6.07 L2 and v L3 chain and L3 produced with IPF v6.10 L2 and v1.1 L3 chain over one day (Figure 28). Observed bias is due to the use of an abaque in the new v1., that induces small differences due to the interpolation in-between the values provided by the abaque. However, the introduced differences remain smaller than 1 cm. Figure 28: Differences of the S3A intercalibrated SWH in the L3 products with respect to the previous version of the processing chain as a function of the significant wave height. V.3.3 January 2018 System v1.2: introduction of AltiKa mission and Improvement of S3A / J3 cross-calibration Calibration and validation of AltiKa mission is detailed in Section II Monitoring and quality control of the v L3 chain production highlighted a small residual bias of the order of 5 cm between Sentinel-3A (IPF v6.07) and Jason-3 as illustrated on Figure 29 (left). Investigation showed that the cross-calibration formula between Sentinel-3A PLRM and Jason-3, determined initially over the [0-12 m] range (see Figure 9, green curve), presented a decrease in accuracy over the [0-6 m] range where most of the wave population lies. Therefore, the cross-calibration linear correction between Sentinel-3A PLRM and Jason-3 is now computed over the [0-6 m] range (see Figure 9, orange curve). The cross-calibration second order polynomial adjustment used to cross-calibrate Sentinel-3A PLRM and SAR (IPF 6.07 version) significant wave heights was also modified to avoid being impacted by the [0-1 m] data, showing large discrepancies compared to the [1-12 m] bias distribution. Initially Page 35/ 38