The Copernicus Sentinel-3A mission was successfully launched in February 2016. It is a multi-instrument mission to measure surface topography, sea- and land-surface temperature, ocean colour and land colour with high-end accuracy and reliability. To reach the data coverage required by the Copernicus service, its twin Sentinel-3B was launched on 25th of April 2018. After 5 months in a tandem flight configuration (Sentinel-3B 30 seconds ahead of Sentinel-3A), Sentinel-3B was moved to its final orbit, the same as Sentinel-3A but shifted by 140°.

The main objective of the Sentinel-3 Surface Topography Mission (STM) constellation over ocean is to provide accurate measurements of sea surface height to support the ocean forecasting systems and climate monitoring applications. To ensure the missions requirements are met, ESA and EUMETSAT have defined the Mission Performance Centre (MPC), a consortium in charge of the qualification and the monitoring of the instrument performances.

In this paper we propose to present the results obtained in the frame of the MPC project focusing on the assessment of Sentinel-3 A and B altimeters performances over Ocean. The SRAL (Sentinel Radar Altimeter) sensors on board Sentinel-3 missions differ from previous conventional pulse limited altimeters by providing observations with the Synthetic Aperture Radar mode (SARM). Along with this processing, an LRM like processing can be applied to retrieve the geophysical parameters as for LRM but with a higher noise (the so-called P-LRM processing). The analyses will be performed over the longest time series available to detect potential strong drifts. Cross comparisons between the two satellites and with other altimeters will be presented to demonstrate the SARM measurements high performances.

Area D - Board 473

Radar altimeter wind and wave data are very important products for ocean wave and atmospheric models. They are used for model validation and for data assimilation (wave height only). The data need validation before they can be used for practical applications. Surface wind speed and significant wave height observed by Sentinel-3A and B (S3A & S3B) radar altimeter (SRAL) in near real time (NRT) have been monitored and validated against the corresponding parameters from the ECMWF Integrated Forecast System (IFS), in-situ buoy and platform instruments and other satellites (specifically: Jason-3, Jason-2, Cryosat-2 and SARAL/AltiKa) including cross-comparison since the availability of such data. S3A and S3B NRT Level 2 STM Marine Product (SR_2_WAT) is routinely obtained from EUMETSAT Copernicus Online Data Access (CODA) are used for this purpose.

Apart from the long-term results of assessment, the assessment during the period when S3A and S3B were orbiting the Earth in tandem (few seconds apart) will be presented. This is particularly important for the validation of Delay-Doppler (SAR) Mode data from S3A with respect to the conventional altimetry Low-bit Rate Mode (LRM) data from S3B (one cycle during the commissioning phase of S3B). It is also important for the cross validation of SAR data from both instruments during the remaining period of the tandem mission.

The NRT availability and timeliness of wind and wave data from both S3A and S3B are monitored since the wave height data are assimilated in the ECMWF wave model. Therefore, the NRT availability and timeliness of wind and wave data will be assessed and presented.

Numerical experiments were conducted to assess the impact of assimilating SWH from S3A NRT data in addition to those from Jason-3, Jason-2, CryoSat-2 and SARAL/AltiKa in the ECMWF third generation wave model ECWAM which is an integrated part of IFS. New set of experiments to assess the impact of assimilating S3B altimeter wave height data will be conducted once the satellite is in its permanent orbit for at least 2-3 months. The results of those two sets of experiments will be presented and discussed.

Area D - Board 474

Usually, the sea surface is characterized by wind sea systems generated by local winds, and swell ocean waves generated by distant weather systems. Although the former are not large enough to disturb the altimetric signal, the swell systems, with larger dominant heights and wavelengths, may play an important role on how it may affect Hs and U10 retrieved by radar altimeter.
The sea state bias (SSB) noise on the altimetric signal, depends mainly on the noise of the significant wave height Hs derived from the altimeter, and its relationship with the backscatter
power σ0 often used to derive the wind speed U10. The methodologies normally adopted for sea SSB estimation are usually developed empirically from non-parametric techniques based on these two geophysical predictors, but it is known that only this information is not sufficient to explain some of the sea surface height (SSH) variability related with sea state and how it can affect the SSB modulation.
Sentinel-1 Level-2 Ocean (OCN) products operating in the nominal acquisition Wave mode (WV) for wave applications, can provide the Ocean Swell spectra (OSW) component in the form of imagettes with a spatial resolution of 20 km by 20 km. The product contains a two-dimensional ocean surface swell spectrum and additional parameters, such as the dominant wave direction, length and height for each wave partition derived from the ocean wave spectra (integrated wave parameters) and from the imagette (image statistics).
This study combines different data sources in the sense of using swell information (height, wavelength and direction) retrieved from Sentinel-1 L2 products and infer the possible impact of these sea state correlators on SSB estimations. The approach is focused on a synergistic approach for the year 2017 using Jason-3, Sentinel-3 and Sentinel-1 sea state related parameters in order to inspect the correlations between the swell systems and the altimeter-derived parameters mostly used as predictors for SSB methodologies. A strong dependence on wave direction in the range of 70º-110º and on wavelenght after the 350 m is found for both altimeter retrievals Hs and σ0, leaking to the standard SSB models used and SSH altimeter estimations.

Area D - Board 475

In the last years, huge efforts have been done by different teams to improve the retracking algorithm for both conventional and Delay Doppler echoes. All the research has been mainly focused on the improvement of the Sea Surface Height estimation, especially to better access the small scales of the ocean topography. However, the retracking algorithm provides, in addition to the sea surface height, an estimation of the wave height which properties can be closely correlated to ocean currents [Ardhuin & al, 2017].
In 2008, a paper was published [Mailhes & al, 2008] on the estimation of Cramer Rao bounds (CRB) for conventional radar altimeter waveforms showing that there was some space for improving the estimation of the sea surface height on the one hand but even more for improving the significant wave height, knowing that the variance of any unbiased estimator of the altimetric parameters is bounded below by its Cramer-Rao bound (CRB).
To improve the retracker performances for LRM measurements, CLS developed and successfully validated a solution called "Adaptive Retracker", implementing a new waveform model and a Nelder Mead optimization method with exact likelihood criterion. Dramatic improvements in the estimation performances over ocean have been shown (respectively 10% and 60% of noise reduction on range and significant wave height).
And consequently, similar developments have been conducted at CLS for Delay Doppler measurements (Cryosat-2 and Sentinel-3A). But, while assessing Cryosat-2 and Sentinel-3A SAR-mode data quality over ocean, some discrepancies between Pseudo-LRM and SAR-mode in range (few cms) and mostly in SWH (10 to 20 cm) have been evidenced especially above 2 m wave height (and in low sea-state conditions too) with a significant wave height dependency. Ensuring continuity of wave height estimates over time (including LRM and SAR altimeters) is crucial for long term climate studies. It has prompted us to investigate these differences and to propose corrections.
This paper aims at presenting a review of the current significant wave height estimation performances with both LRM and SAR altimeters and will propose corrections not only for SAR estimates but also for conventional ones.

Area D - Board 476

The application of extreme value analysis to long duration (30 year) global altimeter and radiometer datasets was undertaken by Takbash et al. (2018). In contrast to previous extreme value analyses of satellite data, the dataset was sufficiently long to enable a Peaks over Threshold (PoT) analysis. When applied to altimeter data for wind speed and significant wave height this analysis produced values consistent with buoy validation data and previous numerical model reanalysis datasets. The spatial distributions produced were also consistent with the model reanalysis data. However, the altimeter data shows much greater fine scale structure for wind speed which is consistent with known tropical cyclone activity. The greater data density provided by radiometer measurements offers the potential to address altimeter under-sampling. However, issues associated with the radiometer’s inability to measure wind speed in heavy rain events appears to create an unacceptable “fair-weather” bias at extreme wind speeds. This renders the radiometer data of wind speed largely unusable for the investigation of wind speed extremes. The study also clearly demonstrates the limitations of the Initial Distribution Method (IDM) for extreme value analysis, which is heavily biased by mean conditions.

Based on the outcomes of this study, the proposed presentation will address the issue of decreasing potential errors and the size of confidence limits on the resulting estimates of extremes when applying the PoT approach to altimeter data. One potential method to address this issue is to increase the amount of data points by using data from adjacent regions. This spatial ensemble of aggregate data is potentially equivalent to a longer duration time series as has been shown previously by Breivik et al. (2013; 2014) for pooled data in the time domain. To define regions for spatial pooling of data, the adjacent areas must satisfy two criteria. Firstly, the areas must be far enough apart that the data points of these areas are independently distributed (i.e. uncorrelated) and not excessively influenced by the same storms. Secondly, the areas must still be representative of a similar wave climate by showing a common parent distribution. To satisfy these two criteria, firstly, data from the ERA-Interim reanalysis are used to determine the spatial coherence of wave height on a global basis. Figure 1 shows correlation ellipses for wave height with correlation coefficients equal or greater than 0.5 shown. These ellipses reflect patterns of both swell and wind direction. Based on these correlation ellipses, spatial coherence between grid squares at the same latitude are assessed. The first criterion is satisfied when the correlation coefficient between these areas is smaller than 0.5. The second criteria that requires the wave climate of these areas to be similar is investigated by determining the relative percentage difference between their monthly means (averaged over the duration records) and their averaged monthly 99th percentile values. We assume the wave climate to be similar when the relative percentage difference is less than 10 % for both the mean and the 99th percentile values. For areas which satisfy the two criteria, data from the uncorrelated areas (2 degree grids) are pooled and Extreme Value Analysis (PoT) is undertaken for the spatial ensemble aggregates. The resulting spatial ensembles have equivalent durations between approximately 60 years and 180 years, depending on the corresponding wave climate zone. The confidence limits for the resulting extreme values from pooled data is calculated, which shows a decrease compared to confidence limits of extremes for each separate area of the composite wave climate region.

The approach enables extreme value estimates to be determined globally from altimeter data with much greater statistical confidence than traditional approaches.


References
Breivik, Ø., O. J. Aarnes, J.-R. Bidlot, A. Carrasco, and Ø. Saetra, 2013: Wave extremes in the northeast Atlantic from ensemble forecasts. J. Climate, 26 (19), 7525-7540, doi: 617 10.1175/jcli-d-12-00738.1.
Breivik, Ø., O. J. Aarnes, S. Abdalla, J.-R. Bidlot, and P. A. Janssen, 2014: Wind and wave extremes over the world oceans from very large ensembles. Geophys. Res. Lett., 41 (14), 614 5122-5131, doi:10.1002/2014gl060997.
Takbash, A., I.R. Young and O. Breivik, 2018: Global Wind Speed and Wave Height Extremes Derived from Long-duration Satellite Records (Draft Version), J. Climate (accepted for publication), doi: 10.1175/ jcli-d-18-0520.1.

Area D - Board 477

Due to the smaller ground footprint and higher spatial resolution of their SAR mode, Sentinel-3A and Sentinel-3B SRAL observations are expected to resolve signals associated with dynamical processes smaller than the large mesoscale eddies (order of 100 km), as well as to have overall improved performance in coastal areas. These two improvements in SAR altimetry observations are assessed in the coastal region of southwest England by comparing Sentinel-3 SAR mode performances against those of Pseudo Low Resolution mode (PLRM). Southwest England is particularly well suited for this task due to the presence of a dense network of observing platforms which provide the in-situ observations against which Sentinel-3 performances are evaluated. The platforms include several moored buoys which collect continuous time-series of metereological and surface ocean variables, as well as an array of HF radars which collects 2-dimensional surface currents and waves on a hourly basis.

Observations from the buoys are used to evaluate Sentinel-3 performance in retrieving significant wave height from the shape of the returning echoes. The buoys include the 16 buoys from the Coastal Channel Observatory, and the E1 buoy from the Western Channel Observatory. The buoys are evenly distributed along the coast of southwest England. Therefore their locations provide a broad range of wave height and swell conditions against which to assess Sentinel-3 SAR observations.

HF radar observations are used to assess both Sentinel-3 significant wave height and, indirectly through across-track surface geostrophic currents, sea surface elevation. Radar observations extend for ~100 km off the northwestern coast of Cornwall, providing a 2-dimensional field which is used to assess Sentinel-3 performance not only on a point-to-point basis but also in terms of reconstructed spatial patterns. Furthermore, the 2-dimensional observations are used to investigate to which degree the accuracy of Sentinel-3 observations is associated with given velocity patterns or swell characteristics.

Area D - Board 478

Since the launch of Sentinel-3 A in February 2016, the data acquired in SAR mode over ocean have opened up a very wide field of investigations. Indeed, the improvement of the along-track resolution and the reduction of the noise of the measurements made the scientific community wondering if ocean phenomena with smaller scales are observable. Nevertheless, before we can interpret all the small scales observations, it is important to fully understand the content of the altimeter signal in SAR mode.
Among the phenomena likely to alter the performance of the measurement, as long as they are not considered in the processing, we can mention the surface motion. Indeed, Doppler altimeters, which use the motion of the platform to improve the resolution of geophysical estimates of the surface, assume that the surface is stationary. In the event that the surface movements are not negligible, the processing assumptions are no longer valid and the final performance is likely to be impacted.
A first analysis has been performed in the context of a CNES/CLS/MIO PhD and resulted in a scientific paper submitted and accepted by IEEE TGRS (O. Boisot et al. 2017). In this work statistical solutions have been developed to estimate the means and standard deviations of the Doppler shift induced by the surface motion. Considering the average values and associated standard deviations, we were able to highlight the possibility of encountering situations where performance is significantly impaired. Unfortunately, it was not possible to deduce the probability of these occurrences or to deduce a deterministic relationship between the sea state and Doppler shifts and hence the impact on the performance.
To complement this work, a new analysis has been performed to accurately assess the impact of sea surface movements on Doppler altimetry performance. The objective is therefore to characterize more precisely the effects of these surface movements on the performance and then analyze possible improvements of the current processing or develop new algorithms to improve the performance. This new analysis is based on a theoretical development allowing to describe the long and short waves time evolution and then to use simulation tools to assess the impact on the performance. Finally, real SAR mode data will be used to validate the theoretical conclusions.
This talk will give an overview of the analysis performed on the time evolution of the short and long wave scales and how this analysis has been used to assess the impact on the delay/Doppler altimeters performances. The results using real SAR mode data will be also presented.

Area D - Board 479

The Baltic Sea, a semi-enclosed and seasonally ice-covered water body, is located in the northern part of Europe where the changes in the wind direction can lead to significant spatial and temporal variations in the wave climate. Variability in waves has a substantial impact on shipping, coastal infrastructure and coastal population. Therefore, understanding the long-term and short-term wave climate make better planning decisions and minimize future hazards and economic loss. Lack of reliable large-scale wave height measurements leads to less knowledge of wave climate variations in the Baltic Sea and the drivers of these variabilities. By recent advances in satellite altimetry, a large number of wave height measurements with great spatial coverage over the Baltic Sea has been collected. This dataset makes it possible to scrutinize variability in the wave climate in great detail. The whole Baltic Sea is covered with multi-mission satellite altimetry observations, from 1990 till 2015 with ~700000 measurements. The data were cross-validated with in-situ measurements and doubtful measurements, such as ice cover, distance from the land and biases between different missions removed. To analyze the spatio-temporal variability of Significant Wave Height (SWH), an Empirical Orthogonal Function (EOF) method is applied to the satellite altimetry wave height data. The main purpose of EOFs is to provide an expansion of the data in a series of functions that separate the spatial and temporal variations. EOF analysis is a robust tool to determine the correlations between the wave climate and climatic indices. A detailed analysis of biases introduced by the temporal gaps and spatial inhomogeneity in the multi-mission satellite altimetry dataset was performed. For this, a simulated dataset following a linear trend of 0.005 m/yr and west-east spatial distribution was created. The results of the EOF analysis were compared for both simulated data without gaps and simulated data with added noise and gaps (in the temporal and spatial domain). These results compared with the outcome of the satellites altimetry and a confidence level is determined. The application of the EOF method illustrated few different modes of climate variability. One mode shows a strong correlation with the North Atlantic Oscillation (NAO) index. We will also discuss possible causes and explanations of other modes of the Baltic Sea wave climate.

Area D - Board 480

Since their launch in April 2014 and April 2016, Sentinel-1A and -1B routinely provide directional swell measurements over open ocean basins, accumulating more than 130 000 swell spectra every month. They contain partitioned directional swell spectra and integral swell parameters for each partition, i.e. significant wave height (SWH), peak period and peak direction. Based on these Level-2 (L2) SAR observations, Level-3 (L3) products are estimated and are delivered in the Copernicus Marine Environment Monitoring Services since July 2018. The motivation for these L3 products is twofold: first, they enable following the swell measurements in consistent swell fields to better monitor swell field space-time structure and evolution, which be of interest to study swell decay for instance; second, they enable rejecting observations that appear as outliers based on the self-consistency of a given swell field. This is a unique approach to better qualify the L2 products for downstream applications. These products, delivered on a daily basis, describe the swell integral parameters evolution along propagation in deep waters, from their storm source until land is reached and away from islands and in the absence of currents. They are often referred to as “fireworks” due to their visual aspect when represented on a world map: evidencing swell generated in localized regions coinciding with strong storms, mostly corresponding to extra-tropical events, from which they propagate over thousands of kilometers across entire ocean basins.

The Level-3 analysis is only applied to swell partitions whose peak wavelength is larger than 200m and whose SWH is larger than 30cm, which rejects approximately 33% of all the swell partitions. Based on this dataset, swell observations are gathered into consistent swell fields and their self-consistency analyzed for all integral parameters. After applying these successive filters, about 45% of all the available swell partitions remain.

The accuracy of the Level-3 products is evaluated and compared to that of the L2 using several methodologies: static and dynamical validation (i.e. considering observation point or propagating the observation respectively) and using comparison against model, buoys and Sentinel-1 self-consistency.

Cross-overs between Sentinel-1A and -B observations are performed to assess the self- consistency of the swell measurements using Level-2 and Level-3 products in various configurations (incidence angle, look angle, satellite mission). Between the L2 and the L3 products, the self-consistency is much improved.

Comparisons against model are performed for L2 products. SWH performances highlight dependences of the wave direction with respect to the sensor azimuth. Also, dependences with respect to the meteorological phenomena are evidenced.

Finally, L3 products are evaluated using triple co-locations with in situ and numerical wave model outputs from WW3. In situ measurements are pre-filtered using the swell measurements self-consistency over time. This inter-comparison is presented for specific storm cases, looking at extra-tropical and tropical cyclones at buoys located at various distances from the swell generation regions. Comparisons show good consistency for major storms close enough from the source. With increasing distance and decreasing energy at the buoy, the Sentinel-1 Level-3 shows over-estimated SWH compared to the model, while the buoy is unable to catch the swell system.

Area D - Board 481

Radar remote sensing is a unique tool for global ocean surface monitoring in many civil applications such as ship detection, search and rescue operations, etc. For these tasks which are related to target identification in a complex environment, sea clutter discrimination is a key issue.
With an imaging radar operating at low spatial resolution the pixel intensity, which is proportional to the backscattered sea surface signal on each radar cell, is known to follow an exponential probability distribution. However, as the resolution increases, the backscattered intensity of sea clutter deviates from this distribution and is better represented by heavy tail distributions such as Pareto, Weibull or K distributions. Now, even though these distributions provide satisfactory fits with the observations using a limited number of tuning parameters (typically, the ”shape” and ”scale” parameter), they are purely empirical and cannot be used for a systematic inversion of the sea state parameters. Nevertheless, there is experimental evidence that the actual intensity distribution depends on sea state. In this work, we propose to establish a first theoretical link between the sea state and the probability density function (pdf) of pixel intensity using a physical scattering model. For the moment, the inversion is limited to the mean square slope (mss) of ocean waves slopes, where only waves larger than the pixel size are taken into account, that is a filtered mss. A strong point of the inversion is that it addresses the distribution of normalized intensities and therefore does not require the absolute calibration of the radar cross section.
In the framework of a Two-Scale model, the intensity distribution can be written as a compound distribution :

p_I(u) = σ_L⁰(θi;s) exp^(−σ_L⁰(θi;s)) p(s) ds

where σ_L⁰ is the radar backscattering cross-section of a rough facet of size L (normalized by its mean value over all facets), θi is the incidence angle, s is the facet slope vector and p(s) the facet slope distribution. This last quantity is assumed to be Gaussian and is solely characterized by the directional filtered mss (say, mssx in the radar line of sight and mssy in the cross direction). The local facet cross section, σ_L⁰, is described by the Bragg scattering model at the local incidence angle. In the end the compound distribution p_I depends only on the resonant spectrum at the Bragg wave number and the facet slopes mss.
It is found that the mssx is the dominant parameter of the compound distribution p_I(u) while the mssy has negligible impact. The inversion of mssx can be performed from the radar image intensities in each polarization (vertical and horizontal) by minimizing the Bhattacharyya distance between the experimental and the theoretical distributions and a very consistent estimator of the mssx can be achieved by combining the results obtained in each polarization. The estimation of the cross direction mssy can only be obtained by changing the radar look direction.

Area D - Board 482

Conventional fitting methods used to retrieve sea surface heights from altimetry waveforms provide the by-product of significant wave heights (SWH). Together with the back scatter coefficient, SWH describes the sea state. However, these algorithms, called retrackers, are not tuned for precise SWH estimation, which leaves them unable to quantify extremes and prone to high noise. In the Sea State Climate Change Initiative Project of the European Space Agency a new improved sea state record is sought with new or improved methods.

For 25 years pulse limited altimeters have observed the ocean, but with the launch of CryoSat-2 in 2010 the technique of Delay-Doppler or SAR altimetry has been established. The two ESA satellites Sentinel-3 A and B, launched 2016 and 2018, also carry a SAR altimeter and are the first missions observing in SAR only. SAR waveforms differ from LRM waveforms in the way they are generated, therefore dedicated algorithms are needed for SAR altimetry.

The slope of the leading edge is used in LRM as a proxy to derive SWH, while the latter also influences the trailing edge in SAR. In this study we investigate the possibility to consider only a portion of the waveform focused around the leading edge for SAR, in order to increase the precision in the open ocean and the amount of retrievable data in the coastal ocean. The functional form is a simplified Brown-Hayne model as in ALES+ (Passaro et al., 2018), but the slope of the trailing edge is estimated as well. The relationship between rising time of the leading edge and SWH has to be revised compared to the LRM case.

In order to establish the relationship between rising time of the leading edge and SWH we simulated SAR waveforms generated with a SAMOSA-like functional form (the model used in the standard products of Cryosat-2 and Sentinel-3) with different wave heights and other parameters. From these simulated waveforms we found a polynomial relationship between rising time and SWH which can be used in the retracker as a correction to estimate directly the SWH. Here, we found a dependency of the SWH estimation and the point target response which has to be considered in the analytical correction as well.

First preliminary results show an improvement of up to 50% for the lower SWH (between 1 and 4 m) compared to the official product, i.e. the SAMOSA approach. For middle high SWH (between 4 and 6 m) the noise is on a comparable level while for high SWH the conventional product has a lower noise.

Figure 1: Sentinel-3 waveform with fitted functional form of subwaveform retracker. By estimating the value of trailing edge slope, the subwaveform retracker is able to fit the trailing edge part of the waveform.

Figure 2: The noise level of SWH. For each SWH the mean of the consecutive differences is calculated.

Figure 3: Relative noise level of SWH of the current version of the new retracker compared to the official (conventional) product.

Area D - Board 483

The CFOSAT (China France Oceanography SATellite) mission is a CNES-CNSA space mission for ocean surface wind and wave observation. The satellite has been successfully launched on October, 29th, 2018 from China. The satellite behaves perfectly well until now with nice scientific acquisitions. For the very first time, wind and waves are observed at the same time and place. These simultaneous observations open new research fields on climate monitoring, sea-air interactions understanding and sea state forecasting.
CFOSAT embarks two payloads: a wave scatterometer, SWIM (French contribution) and a wind scatterometer, SCAT (Chinese contribution). Both are Ku-band real aperture radar taking advantage of different backscattering properties. At high incidence angles (SCAT), the radar signal is sensitive to Bragg scattering, i.e. small surface roughness induced by the wind. At low incidence angles (SWIM), the radar backscattering depends only on large slopes, i.e. the waves.
In this paper, we will present the status on the SWIM instrument and the SWIM products produced by the French ground segment. This status will be provided after 7 months in orbit, at the end of the commissioning phase. At the time of the writing of the abstract (few days after the launch), we are just checking the health of the satellite and the instruments as well as the generation of the first products. We are not thus able to provide quantitative results to the reviewers at this stage. During the commissioning phase, the internal calibration modes of SWIM are tested (internal impulse response, antenna calibration law, reception noise level). The different acquisition modes are also tested (sequence of the six beams and on-board processing options).
The CFOSAT system is composed of the satellite with SWIM and SCAT on-board and two ground segments, one in China (CHOGS) and one in France (FROGS). Both mission centers process all the SWIM and the SCAT data. This paper is relative to the processing performed at FROGS for SWIM only. The CFOSAT data are processed in near real time in the French mission center up to level 2 (CNES) and in differed time from level 2 to level 4 (IFREMER). The products are available on-line on the AVISO+ website.
The main scientific products of SWIM are the 2D wave spectrum at a spatial scale of 70x90 km², the backscattering coefficient profiles per 0.5° step in incidence and 15° step in azimuth and the significant wave height and wind speed from the nadir beam.
The paper will detail the performance of the instrument relative to SNR and backscattering coefficient accuracy, pointing knowledge accuracy. The overall availability of the data will also be discussed (coverage over ocean, distance to coasts, coverage of the continents and ice sheets).
We will also present the results on level 2 products.

Area D - Board 484

The Chinese-French mission CFOSAT was successfully launched in late October 2018. Its aim is to provide the scientific community with remote sensing estimates of Significant Wave Height direction, amplitude and wavelength.
Thanks to CNES- SALP project, CaSyS activities (Systematic Calval for SWIM) enable to validate the performance of this diffusiometer, notably with respect to models and altimetric constellation.
Based on the Level2P CMEMS Wave data set, a comparison of the historical metric to the new dataset is performed, highlighting the potential lacks or complementarity of both techniques to improve the wave estimates for the current mission (S3, J3, J2, AltiKa, C2, but also for elder ones, Envisat, J1…).

Area D - Board 485

The assimilation of directional wave spectra in wave model has showed an efficient way to improve swell directional properties (Aouf et al. 2017). Since the launch of CFOSAT on 29 October 2018 The SWIM instrument provides additional directional wave spectra with an azimuthal cut-off of 70 m, which is significantly better than the ones retrieved from SAR of Sentinel-1. The SWIM instrument provides also the significant wave height at nadir look incidence. The challenge of using both wave spectra from CFOSAT and Sentinel-1 is relevant to improve the sea state forecast. This will also lead to better estimate of momentum and heat fluxes needed for the ocean/atmosphere coupling implemented in the frame of Copernicus Marine Service (CMEMS-MFC).
The goal of this work is to evaluate the impact of the assimilation of CFOSAT wave data during the calibration/validation phase. Several assimilation runs have been performed with sensitivity tests related to the wavelength cut-off of the SWIM instrument. We also performed a combined assimilation of CFOSAT and Sentinel-1 directional wave spectra. The validation of the results is implemented by using independent altimeters and buoys wave data.
The results are discussed in the analysis and forecast period. The persistency of the assimilation of CFOSAT wave data is analyzed during the forecast period. The impact of the assimilation of CFOSAT is analysed for the peak period. The efficiency of the assimilation for storm cases are also investigated.
Further comments and discussions will be developed in the final paper.

Area D - Board 486

The Chinese French Ocean Satellite (CFOSAT) is an innovative space mission dedicated to the observation and monitoring of the ocean sea state and the sea surface vector winds. Recently launched on 29 October 2018, CFOSAT operates two Ku-band rotating radars: the near-nadir Ku-band wave scatterometer (SWIM) and the dual-polarization, moderate incidence angle, Ku-band wind scatterometer (SCAT). Such instrumental configuration provides regular measurements of sea state parameters, including directional wave spectrum, together with simultaneous wind speed and direction. This unique configuration of two sensors operating at two different incidence angle ranges also gives the opportunity to improve the quality of the retrieved parameters by combining both data sources. In particular, this approach can be applied for the improvement of SCAT wind vector retrievals using SWIM nadir observations.
Similarly to other rotating scatterometers, the SCAT measurements have a poor azimuthal diversity of the backscatter (σ° ) measurements in the central (nadir) part of the swath, leading to larger and systematic wind errors in this swath region. At the same time, the SWIM altimeter provides high-quality near-nadir σ° observations (and derived wind speed estimates), which will be used to further constrain the SCAT wind inversion. In the present study, we show the results obtained from an improved wind retrieval algorithm which is based on the inversion of collocated SWIM and SCAT observations.
The simulation of SCAT and SWIM backscatter measurements is performed with the end-to-end simulators developed in the framework of the CFOSAT mission preparation. To ensure correct data synchronization, every instrument simulation is performed for identical orbit parameters and the same model wind field. The standard scatterometer wind vector inversion procedure based on the Maximum Likelihood Estimator (MLE) is extended to accommodate the SWIM collocated measurements. The NSCAT-4 Geophysical Model Function (GMF) is used for SCAT, while the wind-speed only GMF for near-nadir angles developed by Abdalla (2012) and Gourrion et al. (2002) is used for SWIM. Moreover, the 2D-Var ambiguity removal (AR) is also used in the simulation. A preliminary comparative analysis between the combined SCAT and SWIM inversion and the SCAT-only inversion shows that the wind vector errors are significantly lower in the former. The results from a more comprehensive simulation, which includes more realistic backscatter errors and a full wind retrieval process (including AR), will be presented at the time of the conference. Moreover, although CFOSAT is currently in the commissioning phase, a preliminary comparison of both inversion schemes will be tested with real data if the latter become available before the conference.

Area D - Board 487

We decided to assess the differences between the relevant or most commonly used parameterizations for drag coefficient (CD) for the momentum transfer values, especially in the North Atlantic (NA) and the European Arctic (EA). We studied monthly values of air-sea momentum flux resulting from the choice of different drag coefficient parameterizations, adapted them to momentum flux (wind stress) calculations using SAR wind fields, sea-ice masks, as well as integrating procedures. We compared the resulting spreads in momentum flux to global values and values in the tropics, an area of prevailing low winds.

We show that the choice of drag coefficient parameterization can lead to significant differences in resultant momentum flux (or wind stress) values. Comparing the values of momentum flux across the sea surface from the power law parameterization, it showed that in both regions, with low and high winds, the parameterizations specified for all winds speeds has lower values of wind stress than the parameterizations specified for light winds. In the Arctic, the NA, and globally the differences between the wind stress, depend on formula used, are 14 % and they are higher in low winds areas. The parameterizations generally have a decreasing trend in the resultant momentum flux values, with the most recent A12 (Andreas et al., 2012) producing the lowest wind stress values, especially at low winds, resulting in almost 20 % differences in the tropics. The differences can be much larger on monthly scales, up to 29 % in the NA and 36 % in the EA (in months of low winds) and even 50 % locally in the area west of Spitsbergen. For months that have the highest average winds, the percentage differences are smaller (about 7 % everywhere), but because absolute value of the flux are largest for high winds, this 7% discrepancy is also important for air-sea momentum flux values. Since momentum flux is an important parameter in ocean circulation modeling, we believe more research is needed (one aspects that needs more research is the fact that the newest power law parameterization, A12, produces less momentum flux than all the previous ones, especially in lower winds), and the parameterizations used in the models possibly need further work.


Acknowledgements

We would like to express our gratitude to Ed Andreas for inspiring us. His untimely departure is an irreplaceable loss to the air-sea exchange community. We would also like to thank the entire OceanFlux team. This publication was financed with funds from Leading National Research Centre (KNOW) received by the Centre for Polar Studies for the period 2014–2018 and from OceanFlux Greenhouse Gases Evolution, a project funded by the European Space Agency, ESRIN Contract No. 4000112091/14/I-LG.

Area D - Board 488

Local winds and faraway storms ocean waves that can be classified into two categories: wind waves and swell [1]. When wind starts blowing over the ocean, initially short-wavelength waves (capillary waves) are created. As the wind keeps blowing, progressively more energy is added into the system (sea) which generate waves characterized by longer wavelengths. The wave system can be described by a wave spectrum which carries on information about the energy associated with the ocean waves [1]. If the wind blows enough in terms of duration and fetch, a final stage is reached where no more energy can be added by the wind and swell waves generate. The ocean wave spectrum includes both wind waves and swell. The wind-sea part of the spectrum is affected by local wind, while the swell could be generated somewhere else. For a very large fetch and very long wind action time, fully developed sea state conditions are achieved. In this case, the sea is usually defined as stationary or saturated. However, the non-fully developed sea state is the most frequent condition, due to the presence of geographical boundary or any condition that changes the wind speed or the wind direction.
Operationally, in order to design simple and effective spectral-based methods aimed at retrieving geophysical parameters (e.g.; significant wave height, wind, etc.) the fully developed sea state conditions is assumed, i.e.; the wind is supposed to blow for such a long time and over such a long distance that an equilibrium state is reached that depends only on the wind speed [2]. From an operational viewpoint, the assumption of fully-developed sea state conditions simplifies significantly the retrieval of geophysical parameters. In fact, under non-fully sea states conditions, sea surface characteristics that are not directly related to the local wind need to be considered [2]. Hence, it is important to characterize sea state conditions to both design retrieval techniques and to correctly interpret the retrieved parameters.
In this study, a spectral method, which is based on the peculiarities of Synthetic Aperture Radar (SAR) imaging of ocean surface, is proposed to provide a fine spatial resolution inverse wave age map. To the best of our knowledge, no study related to SAR-based estimation of inverse wave age has been conducted. We propose a novel approach that consists of analyzing the departure of autocorrelation function (ACF) measured from the SAR image to its Gaussian fit. The departure is then related to the inverse wave age using a χ2 test.
In particular, the sensitivity of χ2 to inverse wave age (Ω) is discussed by introducing a normalized Ω, ωp = 1.2Ω, which is such that when Ω = 0.84, i.e.; under fully developed sea state conditions, ωp = 1. Preliminary results, obtained processing a large dataset of Sentinel-1A VV polarized Ground Range Detection (GRD) StripMap images collected from November 2014 up to April 2015 in the East Atlantic and off the Hawaii at 10 m resolution, show a good correlation between χ2 estimates and the inverse wave age obtained using European Centre for medium-range weather forecasts (ECMWF) products collocated with the SAR dataset.

[1] D. Ganguly, M.K. Mishra, and P. Chauhan, “Deriving sea-state parameters using RISAT-1 SAR data,” Adv. Space Res., vol. 55, no.1, pp. 83-89, 2015.

[2] A. Guissard, C.Baufays, and P. Sobieski, “Fully and Nonfully Developed Sea Models for Microwave Remote Sensing Applications,” Remote Sens. Environ., vol. 48, pp. 25-38, 1994.

Area D - Board 489

Knowledge about the long-term average wind climate offshore is essential for wind energy planning and other applications in the marine environment where in situ measurements are costly and sparse. Inter-calibration of satellite sensors is essential for achieving the best possible accuracy on wind speed retrievals and resource estimation. We propose a method for inter-calibration of Synthetic Aperture Radar (SAR) observations from different European sensors prior to wind retrieval processing.

Envisat delivered SAR observations during 2002-12 and the two follow-up missions Sentinel-1 A/B were launched in 2014 and 2016, respectively. The Technical University of Denmark performs wind retrieval processing routinely for all three SAR sensors over the Europeans seas using the geophysical model function CMOD5.n. The wind maps are available at https://satwinds.windenergy.dtu.dk/.

Comparisons of the retrieved SAR wind speeds against collocated wind speed data from ocean buys and numerical modeling lead to very different mean biases on the wind speed from sensor to sensor. Likewise, simulations of the Normalized Radar Cross Section (NRCS) from the model wind speeds through application of CMOD5.n in forward mode show residuals with respect to the observed NRCS. These residuals vary according to the SAR sensor and scanning mode and an almost linear increase occurs in the range direction. Further, the residuals vary over the lifetime of the SAR sensors. Based on linear fitting, we calculate corrections of NRCS for each sensor, mode, radar incidence angle, and month. We apply the corrections to the observed NRCS and this leads to a significant reduction of the NRCS residuals, especially for high incidence angles.

The wind retrieval processing is repeated with the corrected NRCS input and we compare once again with buoy observations of the wind speed. For Envisat, the mean bias is reduced from 0.92 m s-1 to only 0.20 m s-1 and the RMSE from 2.37 m s-1 to 1.90 m s-1. For Sentinel-1 A, we see a reduction of the mean bias from 0.16 m s-1 to -0.06 m s-1 and for Sentinel-1 B, the bias is reduced from 0.17 m s-1 to 0.09 m s-1. RMSE is almost unchanged for both Sentinel-1 A and B.There is generally a good agreement between observations from these two sensors except for the first few months of Sentinel-1 A acquisitions obtained during the commissioning phase. These observations are excluded from further analysis.

Improvements on the wind speed accuracy are reflected in our estimations of the wind resource at the ocean buoy locations. For 10 out of the twelve buoys considered in our analysis, we find that differences between the wind power densities estimated from SAR and the buoy wind speeds decrease after inter-calibration. The average deviation from the buoy observations is 20% before and 8% after inter-calibration.

The inter-calibration method presented here can be applied to observations from any kind of SAR sensor. This gives users of the SAR observations the possibility of creating more consistent time series of wind speed and thereby more accurate wind resource estimates. Potentially, such time series could be extended to include SAR observations from third party missions; thereby increasing the sampling density. The inter-calibration method may be refined in the future to also take wind speed and directions as well as the thermal stratification of the atmosphere into account.

This work received funding from the EU H2020 program under grant agreement no. 730030 (CEASELESS project).

Area D - Board 490

The comparison of equivalent neutral winds obtained from (a) four WHOI buoys in the subtropics in the Atlantic and Pacific ocean and (b) scatterometer estimates at those locations reveals a very low root-mean-square difference (RMS) on the order of 0.5-0.7 m/s and a seasonal cycle in the RMS. To investigate this RMS, different buoy wind error sources were examined. Our buoys are particularly well suited to examine two important sources of buoy error: (1) redundant anemometers and a comparison with numerical flow simulations allow us to quantitatively assess flow distortion errors, and (2) one-minute sampling at the buoys allows us to examine the sensitivity of buoy temporal sampling/averaging in the buoy-scatterometer comparisons. The flow distortion can be estimated to up to 5% of the relative difference of the anemometers. A dependence of the error to the measurement position is shown either. Application of this error to the individual anemometer and subsequent comparison with scatterometer estimates show a good agreement. Application of a reasonable time averaging, subtraction of a mean bias, and application of a viscosity correction generally reduces the RMS but cannot explain the seasonal cycle of it.

Area D - Board 491

In the mean wind fields over the Mediterranean Sea produced using the NASA QuikSCAT and the EUMETSAT ASCAT scatterometer data, we noticed signatures of high winds hardly ascribable to geophysical phenomena, located along offshore straight lines and close to coasts. To detect and remove the wind data producing these signatures an outlier detection method, based on the analysis of the regional maxima of the wind speed and on sharp direction turning inside each swath, has been applied to each scatterometer swath. The results show that the location of the wind speed outliers is mainly close to coasts and along the known main ship routes. Statistics from the application of the method to one year of QuikSCAT and ASCAT A and B scatterometer data tells that the rejected winds are 16% and 10% of the total amount of data, with mean values of 7.4 m/s and 6.4 m/s respectively. The amount of data contaminated by ship is large, and should be verified with other methods. Some local artificial feature still remains in the mean wind fields derived from the corrected data set, especially in the areas always congested by ships, as at the entrance of the Suez Canal, again indicating the necessity of further research on the relationships between the presence of ships in the scatterometer cells and the resulting wind, which may depend on the geometry of the measure, on the number, size and position of the ships inside the footprint and on the wind speed regime. The scatterometer wind data producing these signatures should be removed or flagged, because they may affect the geophysical results especially in regional seas and coastal areas. The results found in this paper apply to all the areas where the presence of ships is massive, and therefore are not limited to the Mediterranean Sea. They apply also to all the scatterometer data and not only to QuikSCAT and ASCAT.

Area D - Board 492

Determination of the wind field from SAR in coastal areas

The possibility to get the wind field from the Synthetic Aperture Radar (SAR) images is extremely important in regional seas and coastal areas, only partially covered at present by satellite wind data. Here, the wind fields are often not well reproduced by both global and regional atmospheric models (Accadia and Zecchetto, 2007, Zecchetto and Accadia, 2014), probably because of the interaction between the wind flow and the orography and effects of the local sea water temperature. Furthermore, in small basins the atmospheric model winds are seldom validated, due to the lack of in-situ winds observations. To retrieve the wind field from SAR two things are mandatory: a reliable radar-backscatter versus wind speed algorithm and a suitable estimate of the wind direction.
A method based on the 2D Continuous Wavelet Transform has been developed and applied to SAR images from different satellites (Envisat, COSMO-SkyMed, Radarsat-2 and Sentinel-1A,B), and the results have been compared with the analysis wind fields from the European Centre for Medium-range Weather Forecasts (ECMWF) model, with in situ reports and with scatterometer data when available. The 2D-CWT method provides satisfactory results, both in areas a few kilometres from the coast and offshore. It is reliable as it produces good direction estimates, no matter what the characteristics of the SAR are. Statistics reports a success in the SAR wind direction estimates in 95% of cases (in 83% of cases the SAR-ECMWF wind direction difference is < 20 degree , in 92% < 30 degree) with a mean directional bias < 7 degree . The SAR derived wind directions cannot be said to be validated, as the data available at present cannot be really representative of the wind field in the coastal area.
However, the figures given by SAR winds are highly valuable even not properly validated, providing an independent and unique view of the spatial variability of the wind over the sea, which is possible by using the 2D-CWT method to derive the wind directions. The problem of the wind direction determination from SAR is thus virtually solved by the 2D-CWT method without any external information, and the resulting wind fields can be analysed and compared with those derived using WRF wind directions and with the OWI ESA products.

Area D - Board 493

Scatterometers play an important role in obtaining ocean vector winds or stress. The Advanced Scatterometers (ASCAT) onboard the MetOp satellite series have been generating wind products with reliable quality. They are fixed fan beam scatterometers working at C band, which are less affected by atmospheric rain effects. OSCAT-2 is the rotating pencil beam scatterometer onboard SCATSAT-1 working at Ku band, and rainy conditions bring uncertainties to the wind observations, resulting in rejection flags by the quality control (QC) in rainy areas. For some of those areas, that constitute about 5% in the data sets of OSCAT-2, C band scatterometers may indicate that OSCAT winds are in fact acceptable, from the analysis of the collocated observations of those two type of scatterometers. The QC correction of those products to acceptable quality will increase the effective coverage of winds from OSCAT-2.

However, the differences in observations for those two types of scatterometers are complicated involving differences in Geophysical Model Functions (GMFs) for different frequencies and sampling method of ocean surface due to variance in observing techniques for those two type of scatterometers. The occurrence of rains further causes response differences at C and Ku band due to both atmospheric and ocean surface effects, and make it not possible to resolve a correction model based on physical arguments. By investigating the collocated sets, we attempt to empirically find OSCAT parameters that can predict either consistency or inconsistency/dispersion of the OSCAT and ASCAT wind vector differences.

In our research, the prediction method to OSCAT-2 product quality in those areas is obtained by building up an Artificial Neural Network (ANN). By the established ANN, the differences in ASCAT and OSCAT wind vector observations are associated and expressed through the OSCAT input parameters and allow to effectively accept cases with reasonable ASCAT and OSCAT correspondence.

Further research will be concentrated on the correction accuracy improvement of the ANN, and analysis of the obtained ANN to extract information of details of effective factors causing the QC corrections of the OSCAT observations. The results are expected to be beneficial to the QC of all past and future pencil-beam scatterometers.

Area D - Board 494