The ocean circulation in the northwestern Mediterranean Sea exhibits widespread mesoscale dynamics, with strongest values along the Northern Current which flows westwards along the French coast following the continental slope (Millot, 1999; Guihou et al., 2013). Observing the mesoscale variability is critical in this region since it plays a key role in the coupled ocean–atmospheric system that can lead to extreme precipitation events (Lebeaupin Brossier et al., 2015). Horizontal currents stirred by the mesoscales are important in the dispersion of pollutants and the monitoring of marine ecosystems. The vertical transport of heat, salt and nutrients is strongly driven by the smaller-scale dynamics, in the fronts and filaments surrounding these mesoscale eddies, and within the deep convection cells that form in the Gulf of Lyons in winter–spring (Herrmann et al., 2008).

Compared to other current systems at similar latitudes such as the Gulf Stream, the mesoscale variability in the northwestern Mediterranean Sea has a small Rossby radius of 5–15 km, varying seasonally with the stratification (Grilli and Pinardi, 1998). This makes the ocean dynamics of this region particularly difficult to observe and monitor. The surface mesoscale characteristics have been studied with satellite sea surface temperature (SST) and ocean colour data in clear-sky conditions (Robinson, 2010), but the mesoscale variability is often hidden in winter by clouds and in summer under the more homogenous warm surface layer. Numerical modelling studies are improving in resolution and in their internal physics to allow a better representation of the mesoscale variability (e.g. Herrmann et al., 2008), although these models need to be validated against observations.

In the global ocean, mapped satellite altimeter products have allowed unprecedented advances in understanding the mesoscale eddy variability and characteristics (Chelton et al., 2011). Altimetry measures sea surface height (SSH) that responds to mass and density changes over the entire water column, and as such, altimetry is the only satellite observation that can detect deep ocean changes. Deep-reaching mesoscale eddies can be tracked over many seasons or years (e.g. Morrow et al., 2004; Chelton et al., 2011), even if their surface signature disappears through air–sea interactions so that they become undetectable in satellite imagery. Although regional altimeter maps have been constructed with improved resolution and spatial scales adapted for the Mediterranean Sea (e.g. Pujol and Larnicol, 2005), the spacing between ground tracks still limits our ability to monitor scales less than 150 km wavelength (or 75 km diameter features) (Pascual et al., 2006). Thus we can only detect the larger mesoscale structures, missing most of the typical Rossby radius dynamics in the Mediterranean Sea.

Along-track altimeter data are able to detect finer scales than the mapped altimeter data, but the spatial scales we can resolve are still limited by the altimeter noise, the accuracy of the corrections and the processing methodology. However, over the last 5 years, there has been great progress in improving the quality of along-track satellite altimeter data for ocean studies. Of the three missions currently flying in the altimeter constellation, Jason-2 in Ku-band (launched in 2008) has benefitted from continually refined algorithms and corrections, and new waveform retrackers that allow more data points to be collected close to the coast and islands, and more stable performance with lower noise over the oceans (Dibarboure et al., 2011). SARAL/AltiKa (launched in 2013) was designed to have a smaller footprint and lower noise over all surfaces, due to the choice of antenna pattern, Ka-band frequency and its lower altitude (Verron et al., 2015). CryoSat-2 (launched in 2010) is primarily a cryosphere mission and not planned for ocean observations. Yet over the last years, considerable efforts have been made by the ESA SAMOSA project (Ray et al., 2015) and the CNES Cryosat-2 Processing Prototype (CPP) project (Boy et al., 2017) in collaboration with oceanographers to improve the waveform retracking over the ocean and provide adequate corrections for ocean observations. CryoSat-2 is in low-resolution mode over most of the global ocean but has synthetic aperture radar (SAR) mode observations available over a few regions, including the Mediterranean Sea, with improved along-track sampling down to 300 m and reduced noise. However, certain ocean corrections are less accurate than on Jason-2 or SARAL, including the radiometer correction and the mean sea surface estimate, since CryoSat-2 is on a geodetic orbit. These three altimeter missions with different technologies and data processing will provide an ideal data set to test the improved observational capabilities in the NW Mediterranean Sea.

Previous studies have analysed the altimetric capabilities in the NW Mediterranean Sea from conventional along-track data (Bouffard et al., 2008, 2011; Birol and Delebecque, 2014; Birol and Nino, 2015), including using seasonal averaging to reduce the noise for Jason but maintaining along-track resolution (Birol et al., 2010). Here we will take a different approach, in order to measure the altimetric signal-to-noise ratio statistically in the different seasons. We will calculate along-track sea level anomaly (SLA) spectra (e.g. Fu, 1983), which allows us to observe the SLA spectral energy at different wavelengths, and also the time-averaged spectral noise at small wavelengths. In terms of signal, the spectral energy of SLA is higher at longer wavelengths, and lower at small wavelengths, and geostrophic turbulence theory involves a cascade of energy from the larger to smaller scales, leading to a steep spectral slope in wavenumber space. When spectra are averaged (over different ground tracks in a region and/or over time along the same ground track), the random altimeter noise averages out to create a flat spectral noise floor in the 1 Hz data. This spectral noise level then defines our altimeter noise. The intersection of this noise floor with the spectral slope will define the limit of the observable wavelengths, where the signal-to-noise ratio is statistically greater than 1.

Following Xu and Fu (2012) we will remove the spectral noise from the spectra before calculating the spectral slope, to improve the slope estimate and have more precise observational limits. This technique has been applied to the global altimeter data sets, for Jason-1 by Xu and Fu (2012) and for Jason-2, SARAL and CryoSat-2 by Dufau et al. (2016). Their results showed considerable geographical variations in the spectral slope, noise levels and mesoscale resolution (Xu and Fu, 2012), and strong seasonal variations in the noise level and the mesoscale observing capabilities (Dufau et al., 2016). Neither study included the smaller Mediterranean Sea region, due to the limited spatial coverage in this regional sea. In our analysis, we will concentrate on tracks having at least 200 km length.

These studies calculated their spectral slopes over a fixed “mesoscale” band from 70 to 250 km wavelength. The Mediterranean Sea, which is dominated by smaller dynamical structures, may have different spectral energy and spectral slopes in this band compared to open-ocean regions. The surface sea-state conditions are also dominated by short wind waves and less by long swell, which may impact on the radar altimeter's noise level. Both of these features will be considered in the first section of this paper. We aim to investigate the noise levels for the most recent altimeter missions, estimated from their spectral noise level in the Mediterranean Sea. We will revisit the appropriate filtering to be applied to remove the noise in different seasons. We will then consider what scales of ocean dynamics can be observed today in the Mediterranean Sea with along-track altimetry and investigate how much of the seasonal dynamical signal is observable above the seasonal noise.

In the second part of this paper, we will use a complementary approach and focus on the observation of individual features using a combination of altimetry and a limited number of glider sections and 2 years of high-frequency (HF) radar observations filtered at similar scales. We will examine whether the ocean scales observable with altimetry are also captured by the co-located in situ data. Glider–altimetry comparisons have been used for previous altimetry missions in the NW Mediterranean Sea (e.g. Bouffard et al., 2010) but not for the three most recent missions. For the glider comparison, we only have a limited number of historical co-located sections, and so gliders were deployed specifically along altimetric tracks for each of the three missions, under different mesoscale conditions. For the HF radar, we will use a HF radar site near Toulon, as part of the MOOSE observational array (Quentin et al., 2013), with an offshore extent of 25–75 km from the coast. We will discuss the strengths and limits of the different measurement systems' observation in the coastal band.

Spectral analyses are performed on each of the three altimeter missions, with their tracks shown in Fig. 1. Only data more than 50 km from the coast are analysed to avoid the increased errors in the coastal zone. Each track and cycle is then selected along a common segment of 200 km. This segment length was chosen to allow a large number of altimeter segments in different regions in between the numerous islands and to be more than 50 km from the coast, to avoid the increased errors in the coastal altimeter data. This segment length is also long enough to well resolve the dominant scales (Rossby radius of 5–15 km). Missing data are a problem for a stable spectral analysis. If fewer than three consecutive 1 Hz points are missing (20 km), the data are linearly interpolated; if a larger gap is present the cycle is eliminated from the analysis. Tracks passing over large islands are thus eliminated (see Fig. 1). Wavenumber spectral analysis is then performed by Fourier transform on the ensemble of the remaining segments for each mission (see Table 1). The cycles are averaged in wavenumber space for the entire period and for each season.

An example of the power spectral density (PSD) of SLA averaged for all of the Jason-2 data in the NW Mediterranean Sea over the period 2008–2015 is shown as the black curve in Fig. 2. The PSD is high at longer wavelengths (> 300 km). There is a cascade of energy over the mesoscale range from 50 to 300 km, but the spectra become whiter at small wavelengths (i.e. less than 50 km), where the weaker ocean energy is hidden by the stronger instrument and geophysical noise.

In the following seasonal analyses, the noise level will be calculated as a constant PSD value estimated between 12 and 25 km wavelength, as in Dufau et al. (2016) (e.g. black horizontal dashed line, Fig. 2).

Following the global studies made by Xu and Fu (2012) and Dufau et al. (2016), we then subtract this statistically stable noise level from the mean spectral curve, to obtain an unbiased spectral estimate corrected for the noise (red solid line curve, Fig. 2). The spectral slope of this unbiased estimate is steeper over the mesoscale range and corresponds to a k-2.5 slope and the SLA PSD cascade continues more smoothly down to smaller wavelengths.

We define the mesoscale observability limit as the wavelength corresponding to the intersection of the spectral slope and the noise level, where the signal-to-noise ratio is greater than 1. This is a statistical representation of the average ocean and noise conditions over the entire period and over the entire region analysed. In some local cases, smaller energetic structures may still be observable above the altimetric noise. However in the following results, we will discuss this regional statistical approach.

The mean spectra for the three altimeter missions over the NW Mediterranean Sea are shown in Fig. 3a for the 200 km segment tracks in Fig. 1 and over the 13-month common data period from 1 April 2013 to 30 April 2014. The unbiased estimate with the noise removed is in Fig. 3b. Recall that the space–time samplings of the three missions are different, and as such they may capture different dynamics at different regions. So we do not expect the spectra to be perfectly aligned. More distinctive are the different noise levels between 15 and 100 km wavelength. Jason-2 has the highest noise level in this region, followed by CryoSat-2 in SAR mode. SARAL/AltiKa in Ka-band exhibits the lowest noise of all.

When a constant noise level is removed from each spectral PSD, the spectral slopes line up surprisingly well, given the different space–time sampling of the three missions over this 13-month period. The spectral slope is again around k-2.5 from a fit to the unbiased spectra over the wavelength range from 50 to 200 km. These spectral slopes in the offshore regions of the Mediterranean Sea are quite shallow compared to the k-5 slopes expected for quasi-geostrophic theory (Stammer, 1997). The reason for this needs further investigation, but smaller slopes are also characteristic of open-ocean low-eddy-energy regions (Xu and Fu, 2012). For the Mediterranean Sea, the dominant mesoscale energy at small Rossby radius scales tends to flatten the spectra, but internal waves or mean sea surface errors in the CryoSat-2 data could also contribute to higher SSH energy at small scales and flatter spectra (Dufau et al., 2016).

The fact that the CryoSat-2 1 Hz data in SAR mode had a higher noise level than SARAL/AltiKa was unexpected. We verified that the CryoSat-2 20 Hz data were consistent with the 1 Hz averages, so this is not an averaging problem. The CryoSat-2 20 Hz SAR mode does exhibit a spectral hump for this region and time period that was not present in other regions with SAR data (Agulhas or tropical Pacific; S. Labroue, personal communication, 2016). This warrants further analysis of the particular surface roughness conditions occurring in the NW Mediterranean during this year, and further expertise in SAR processing for the Mediterranean conditions is needed. These results reinforce the very low noise level associated with the 40 Hz Ka-band SARAL data, averaged here to 1 Hz.

Seasonal spectra were also calculated from the longest time series possible, i.e. over 6.5 years for Jason-2 data, over 22 months for SARAL/AltiKa, and for the shorter 13-month period for CryoSat-2 (see Table 1). The spectral noise floor levels for the seasonal analyses are shown in Fig. 4a. Note the spectral units are in m2 cpkm-1, where cpkm refers to cycles per km. Jason-2 and SARAL/AltiKa show a large seasonal variability in their noise levels, with highest noise levels in winter (1.2 × 10-3 m2 cpkm-1) and then autumn, due to the high sea-state roughness in these months from the stronger wind-wave conditions which increases the spectral SLA “hump” at wavelengths from 30 to 70 km (Dibarboure et al., 2014). In summer, the Jason-2 noise level is only 0.8 × 10-3 m2 cpkm-1, but this is still higher than the noise floor in any season for the SARAL or CryoSat-2 missions. SARAL with its small footprint has the lowest noise levels but has strong seasonal variability, with values ranging from a low 0.3 × 10-3 m2 cpkm-1 in summer to 0.7 × 10-3 m2 cpkm-1 in winter. The CryoSat-2 SAR mode shows very stable background noise levels over this 1-year record, varying between 0.6 and 0.8 × 10-3 m2 cpkm-1. The reasons for this stable seasonal noise level are not yet known. However CryoSat-2 has a long repeat cycle (369 days), so different geographical regions are sampled in different seasons; there may be strong interannual variations in the wind-wave conditions that merit more detailed investigation. The additional mean SSH errors introduced due to the non-repeating track will also impact the CryoSat-2 spectra over all seasons.

Figure 4b shows the observational limits for each altimeter mission by season. Clearly, the background noise is not the only limiting factor on the scales of mesoscale energy that we can observe. The SLA energy at low wavelengths also varies from one season to another. In winter, when the mixed layers are deepest and energetic deep convection cells occur in the NW Mediterranean Sea (e.g. Herrmann et al., 2008), all of the altimeter missions can observe wavelengths down to 40–50 km (individual features of 20–25 km). In summer when the submesoscales are weaker, SARAL can detect ocean scales down to 35 km wavelength, whereas the higher noise from Jason-2 and CryoSat-2 blocks the observation of scales less than 50–55 km. This characteristic was also noted in the global analysis of Dufau et al. (2016). Unfortunately in winter, when we would like to observe the smaller energetic submesoscales, all of the radar altimeters observe higher noise levels associated with the higher wind-wave field.

The previous section highlighted that the altimetric noise was effectively masking the smaller-scale SLA signals in the along-track data. The smallest scales observable with a signal-to-noise ratio greater than 1 will vary from one altimeter mission to another and seasonally. Statistically, we cannot observe structures less than 35–45 km wavelength with SARAL, or 50–60 km wavelength with the higher noise of Jason-2. However, individual energetic features may be revealed above the statistical noise. We will explore this with a series of co-located along-track altimeter–glider sections and compare the vertical structure observed by the gliders with their steric height and geostrophic velocities.

In this section, the filtering of the along-track altimetry data is based on the standard Loess filtering applied to the CTOH coastal processed data (Birol et al., 2010; Birol and Nino, 2015). For each glider–altimeter comparison, the first estimate of the along-track altimeter filtering scales was based on the seasonal spectral analysis results for each altimeter mission (see Sect. 3). Other cut-off frequencies around this seasonal statistical value were also tested. The filter which gave the best results in terms of glider–altimeter correlation coefficient and which had the lowest cut-off wavelength was then chosen. The altimeter filter values are given in Table 2.

One should bear in mind that the glider steric height and geostrophic velocities (with or without their surface drift adjustment) will observe different dynamics from the altimetric sea level and geostrophic velocity anomalies. The steric height calculated from gliders represents the upper ocean baroclinic component due to the density anomalies above 1000 m depth. Altimetric SLAs include the full-depth baroclinic motions and the barotropic component, and the barotropic flow may be quite active in the NW Mediterranean Sea, in particular near the shelf break and slope (F. Lyard, personal communication, 2016). When the glider “surface drift” is added to the glider geostrophic currents relative to 1000 m, this may partially correct for the missing barotropic component. Altimetry may also include other SLA signals, such as from internal tides or internal waves, which contribute as errors in the geostrophic velocity calculation (although tides are small in the Mediterranean Sea). In addition, the altimetric SLAs have the mean ocean circulation removed, whereas the gliders provide the total upper ocean baroclinic flow. For consistency, the mean dynamic topography and mean geostrophic velocities derived from Rio et al. (2014) are added to the altimetric data for this comparison. The third main difference is the time taken to make a section over 100 to 300 km. The altimeter makes a “snapshot” of the section as it passes at 7 km s-1 (200 km in 30 s) whereas the glider moves at 0.35 m s-1 (200 km in 6.5 days). We will see that slow-moving structures may be well-sampled by both; rapidly evolving smaller-scale structures are harder to co-locate.

One crucial point is that the gliders have their own noise and also measure HF ageostrophic ocean structures that will not be observable with altimetry. Figure 5 shows a vertical temperature section over the upper 200 m from the glider Milou along the SARAL altimeter track 57 from 27 October to 3 November 2014. Figure 5b shows the very small-scale signals in the upper ocean temperature structure along this 164 km long section. These may be associated with noise in the glider heading or from the processing steps, or from internal waves or rapid submesoscale structures. To remove these scales, we have applied a recursive Butterworth second-order along-track filter to the density data, before calculating the steric height or geostrophic anomalies, with a filter cut-off at 30 km wavelength, designed to retain the typical Rossby radius scales of 10–15 km in the NW Mediterranean Sea. This filtering step was recommended from previous glider studies (e.g. Durand et al., 2016). An example of the filter applied to the same temperature section is shown in Fig. 5c. Similar filtering is applied to the different glider sections presented below.

Ten glider sections are available, co-located with altimeter tracks (details given in Table 2). Here we present three glider track sections along different altimeter mission tracks.

HF radar data provide an additional observation of the oceanic surface currents. In comparison to the geostrophic component of the flow obtained with altimetry and gliders, HF radars measure the total surface current, due to balanced geostrophic and unbalanced ageostrophic currents (wind-driven, inertial, tidal currents, etc.). The daily data set we used has been processed to remove the HF tides and inertial currents, retaining the geostrophic and wind-driven currents. Figure 10 shows an example of the HF radar total currents for one date, 20 October 2013 near Toulon, with the two coastal radar locations marked. The presence of the strong Northern Current is clearly visible in the 2-D HF radar current vectors, with a central jet only 10 km wide, the current spanning 20 km to its edges. This is clearly below the statistical observability limits from the spectral analysis of the three altimeter missions. The offshore extent of the HF radar data is from 25 to 75 km from the coast, which extends into the coastal band that was excluded from our spectral analysis, as it has frequently “noisy” altimeter data and corrections. The small spatial coverage of the HF radar means that no Jason-2 data cross this region, although we have one SARAL track passing through the centre (Fig. 10) and a number of non-repeating CryoSat-2 tracks. The angle of the SARAL track shown in Fig. 10 is such that the cross-track geostrophic currents are mainly orientated in the principal direction of the Northern Current. For this date (20 October 2013), the amplitude of the HF radar currents, projected in the altimetric cross-track direction (in red), is similar to the SARAL cross-track currents (in black), reaching 0.7–0.8 m s-1 within the Northern Current. Further offshore, the HF radar currents decrease gradually whereas the geostrophic altimetric currents are much weaker outside of the jet. The presence of ageostrophic currents in the HF radar data could contribute to this difference. Our statistical estimate of the spatial observability of SARAL observations in autumn is around 35 km wavelength (Sect. 3), representing feature structures across the current of around 17 km. Clearly at these scales, the 20 km wide Northern Current can be observed by the SARAL altimeter.

The advantage of the HF radar data set is its daily 2-D coverage at fine resolution, so we should not have the space–time offsets in the sampling of small-scale features that plagued the glider–altimeter comparisons. The disadvantage is that altimeter data in the last 10–50 km from the coast are noisy, and the ageostrophic wind-driven component of the HF radar surface currents can be strong here, in the region with strong mistral winds.

We have compared the observability of these near-shore currents with the finer-resolution SARAL altimeter time series, filtered at 35 km (see Sect. 3). SARAL data are available along this track every 35 days, and Fig. 11 shows the 18-month time series of cross-track surface velocities from the HF radar. The upper panel shows the full time series of HF radar currents projected perpendicular to the altimeter track; the middle panel shows the HF radar currents sampled at the same dates as the SARAL altimeter passes, and spatially sampled at 7 km as for the 1 Hz altimeter data. The bottom panel shows the SARAL 1 Hz geostrophic currents (mean and anomalies), filtered at 35 km. SARAL clearly detects more of the offshore return flow than the HF radar can but covers a similar data range as the HF radar to the coast. Along-track correlations of the HF radar and altimetric currents for this cross-track velocity component are between 0.7 and 0.9 for these 16 tracks, except for four dates, where the correlations drop below 0.5. The RMSE between the cross-track HF radar current amplitudes and the SARAL current amplitudes is shown in Fig. 12. Dates with low correlations (< 0.5) are marked with the vertical dashed line, and these have a higher RMSE. The RMSE is generally lower in the summer months when the wind is lower and increases in winter.

Wind forcing of the ageostrophic currents may explain part of the difference. If we consider the daily time series of HF radar data (Fig. 11a) and extract the outliers in cross-track velocity having > 1σ standard deviation from the mean, we find that these outliers are correlated at 0.84 with the cross-track wind at the same date (not shown). For the dates with weak correlations, wind may play a role for one date (December 2013), but the other dates have relatively low wind. The differences with SARAL are often associated with 10 km wide structures and close to the coast. This could be due to errors in either measurement system (e.g. for SARAL: the nearshore wave height bias, wet tropospheric corrections, mean sea surface errors) but also from rapid events that are detected by the altimeter 8 s “snapshot” but viewed differently with the HF radar 1-day averages (rapid meander, internal waves, etc.). Planned future analysis of the higher-frequency radar data and the 40 Hz altimeter data with appropriate filtering may help elucidate some of these differences.

The along-track altimeter spectral analysis allows us to estimate the mean dynamical scales that can be observed today with different altimeter technology and associated processing, and in different seasons. In winter, when the mixed layers are deepest and the submesoscale is energetic, all of the altimeter missions can observe wavelengths down to 40–50 km (individual feature diameters of 20–25 km). In summer when the submesoscales are weaker, SARAL can detect ocean scales down to 35 km wavelength, whereas the higher noise from Jason-2 and CryoSat-2 blocks the observation of scales less than 50–55 km wavelength.

This is a statistical view. There are limits in applying this too assiduously, especially as these statistics are calculated from relatively short records for SARAL, and only 13 months of reprocessed SAR data for CryoSat-2. We chose to analyse the longest time series possible for the seasonal calculations since the records are relatively short. However, entire years should be analysed to remove any sampling biases in these statistics. Given the long repeat time for CryoSat-2, we also measure different geographical regions in each season, which can introduce biases in our basin-scale averages. Interannual variations also occur in the dynamics in response to interannual atmospheric changes, which can lead to different deep convection events from one season to another (Adloff et al., 2015). Analysing a longer time series of SARAL and CryoSat data should improve the significance of these early results.

One application of this type of analysis is to improve the altimetric data post-processing to be adapted to the regional conditions. Today, along-track filtering is applied in a similar way to all altimeter missions to reduce the instrument and geophysical noise. Since consecutive altimeter points are laid down spatially, data are filtered spatially along the track to reduce this noise. Standard filtering in the AVISO along-track products DT2010 ranges from 55 km wavelength at high latitudes to around 250 km in the tropics (Dibarboure et al., 2011). The new AVISO products DT2014 apply lower along-track smoothing at 65 km wavelength, globally and for all missions (Pujol et al., 2016). This study suggests that the along-track filtering may be tuned in a regional study to be better adapted to the local dynamics and noise conditions. Thus in the NW Mediterranean Sea, filtering of Jason-2 data could vary seasonally from 50 km in winter to 60 km in autumn and spring (or a conservative 60 km year-round). SARAL could have a finer-scale along-track filtering applied, to retain wavelengths greater than 35 km in summer–autumn and 45 km in winter. A filter cutoff of 50 km year-round could be suitable for CryoSat-2. Knowing how this statistical signal-to-noise ratio varies from one mission to another, and seasonally, is very useful for regional applications, for local process studies or for data assimilation.

The in situ validation remains very limited in space and time and did not allow us to confirm whether these smaller scales are realistic ocean features. For the glider comparison with SARAL, small-scale structures were detected by both systems, but their rapid movement prevented us from giving a precise along-track co-location except for the short scales close to the temporal crossing point. Indeed, for advective dynamics to be resolved correctly, they should conform to the Friedrichs–Lewy condition, i.e. UΔt / Δx < 1. If we follow small structures with typical advection speeds of U = 0.3 m s-1 (typical of the Northern Current), then we need time differences, Δt, of less than 1.35 days to resolve the smaller SARAL wavelengths at 35 km, and within 2 days for the Jason-2 and CryoSat-2 data to resolve 50 km wavelength structures. With the slow-moving gliders, we can only cover 30 km per day, and so our along-track intercomparisons should be limited to the ±30 km around the altimeter–glider crossing point. This places a very strong constraint on our in situ validation.

The SARAL intercomparison with the Toulon HF radar data was quite promising. Despite the apparent nearshore errors in the SARAL data, and the periods with strong wind-driven currents, the correlation between the SARAL geostrophic currents and HF radar total currents remained high. The position of the Toulon HF radar helps, as the observations are centred on the Northern Current, in a region where the current is strongly steered by bathymetry, and the geostrophic component is dominant. This example indicates that a strong coastal current, with a high signal-to-noise ratio, can be detected by satellite altimetry, even at 20 km from the coast. Improvements are still needed to reduce the altimetric errors in the nearshore region, and to compare the CryoSat-2 SAR current observations with the HF radar data. This good intercomparison suggests that HF radar data may be combined with altimetry to extend the observations (duration and offshore extent) of the Northern Current and its recirculation near Toulon.

Another potential way to cross-validate the feature scales observed by the different altimeter missions is to use the crossover points between different missions. Figure 1 shows that there are many crossover points during this analysis period, especially from CryoSat-2 on its long-repeat 369-day orbit and even from Jason-1, which moved into a long-repeat 406-day geodetic orbit from April 2012 to 1 July 2013. Our analyses of the small, fast-moving features in this paper indicate that we really need crossover measurements overlapping within 1–2 days to capture these fine-scale features. These multi-altimeter overlapping passes are also interesting for the missions on a similar inclination, since their overlapping sections can be quite long. For example, SARAL and CryoSat may have long overlapping sections with a time difference of less than 2 days (see Fig. 1). Similar long sections may be available from the Jason-1 geodetic mission and Jason-2. At present, we are developing the code to calculate the crossovers from multi-satellite passes and select the passes based on their time differences. This analysis will be performed as part of our ongoing work in this region.

For the future altimetric missions, finer spatial sampling and lower noise levels should continue, with Sentinel-3 in global SAR mode launched in early 2016, and SWOT providing 2-D interferometric SAR heights and images and an order of magnitude lower noise in 2021. Similar wavenumber spectral analysis techniques could be applied to estimate the noise levels and observable spatial scales with these new missions. This study illustrates that the difficulties in setting up an adequate in situ validation for the small-scale, rapidly evolving dynamics will remain a challenge to resolve in the future.

Altimeter data: the unfiltered along-track Jason-2 and SARAL altimeter SLA data sets are available from the AVISO website (http://aviso.altimetry.fr/) and the CMEMS website (http://marine.copernicus.eu/). The unfiltered along-track CryoSat-2 CPP data are an experimental product provided by the CNES. These Level 2 (GDR) input data are provided by CNES, ESA, and NASA. CryoSat SLAs are calculated relative to a gridded mean sea surface (MSS_CLS2011, http://www.aviso.altimetry.fr/en/data/products/auxiliary-products/mss.html).

Glider data are available as part of the MOOSE project (http://www.moose-network.fr/gliders).

HF radar data are also available as part of the MOOSE observing system (http://hfradar.univ-tln.fr/HFRADAR).