Agencies in charge of SAR operations, marine pollution response and maritime traffic control are among the most significantly targeted users of reliable met-ocean information. Access to multi-platform quality-controlled near-real-time met-ocean observations and high-resolution forecasts available for their specific areas of responsibility for marine SAR, assigned by the IMO (International Maritime Organization), is essential for them to support emergency response missions. Wind, wave and surface current observations and forecasts are needed to be seamlessly integrated into their SAR emergency tools in order to predict the trajectory of a drifting target for determining the optimal search region. In the sphere of maritime safety, HFRs have the great advantage of providing high-spatiotemporal-resolution surface currents in wide coastal areas very close to the coastline when HFR gap-filling methods are applied (listed in Sect. 2.1.3) and where most of the SAR incidents occur (as shown by Fig. 1), as the review of the location of the SAR incidents from five countries (i.e., Croatia, France, Italy, Slovenia and Spain) along 2019 and from Malta along 2020 clearly shows. i.

Croatia. A total of 612 SAR interventions were registered in 2019, 389 of which are SAR interventions and 223 are MEDEVAC ones (i.e., actions related to transportation of injured or sick persons). Most of these incidents occur during summer from June to September and over 98 % in inner and territorial waters. Coastal waters from Croatia are operationally monitored by the HFR-SPLIT WERA Radar System, consisting of two WERA HFR sites (Ražanj and Stončica) in the eastern part of the eastern mid-Adriatic basin. The HFR-NASCUM system is a historical network located in the eastern part of the Gulf of Venice. All four sites were used to build NEURAL project short-term predictions (described in Sect. 2.1.3).

Aiming to improve the applicability of this model assessment methodology for SAR operations in coastal areas, Révelard et al. (2021) also analyzed the SS sensitivity to different forecast horizons and showed that in coastal regions (i) the SS is sensitive to the forecast time, i.e., the longer the forecast (i.e., 72 h), the higher the SS value, due to the high variability of the surface currents; and (ii) a shorter forecast time (e.g., 6 h), consistent with the duration of the search that maximizes survivors in SAR missions, is therefore more appropriate. In addition, they have shown that, whereas the original definition of the skill score from Liu and Weisberg (2011) is correct for analyzing its spatiotemporal distribution, the use of a novel skill score (SS∗) is recommended to assess the average model performance over an area of interest and along a specified period to avoid biased conclusions. The advantage of the SS∗ is that, despite its similar formulation to the SS defined by Liu and Weisberg (2011), SS∗ does not impose negative values to zero, allowing us to obtain a correct average, as in Fig. 2. However, they clarified that only SS∗ values >0.5 should be interpreted as good agreement between HFR surface current observations and model outputs.

A further academic study of the value of quality-controlled HFR observations in SAR operations was the recent case of a person lost at sea in the northern Adriatic during a Sirocco storm on 29 October 2018. In this case, HFR-NAdr observations were employed for hindcasting and survivor's drift trajectory verification (Ličer et al., 2020). Figure 3 depicts Lagrangian drifter dispersal computed from modeled surface winds (the dominant contribution to the drift in this case) and HFR surface currents from the HFR-NAdr network in the Gulf of Trieste after this accident.

Even though in this case part of the survivor's trajectory outside the HFR-NAdr domain had to be inferred from extrapolated currents, such HFR-based nowcasting products would have been valuable during this and similar rescue attempts. However, since HFR data arrive in near-real time, some sort of model-based extension of their prediction horizon is necessary before they can be used for operational nowcasting. One possible solution is data assimilation of HFR data into a numerical model (see Sect. 2.1.2.), followed by a forecasting time window. An alternative and numerically less demanding option that has gained ground in recent years is the machine-learning approach whereby a neural network model is trained on past data and then used to create short-term predictions of surface currents, as addressed in Sect. 2.1.3.

This section addresses one of the main interests and needs of end users of operational oceanography information: users want to be able to have confidence in modeled data and they need to know how good they are. Addressing end-user overarching concerns, model assessment, essentially built upon comparison to observations, is crucial to evaluate the quality of the diversity of modeling products available in a systematic and long-term routine manner and to inform users about their usefulness for a given application.

For this reason, also seeking to strengthen end-user loyalty, the validation of operational ocean forecasting systems against independent measurements constitutes a core activity in operational oceanography (Hernández et al., 2015) since it aids in (i) inferring the relative strengths and weaknesses in the modeling of several key physical processes, (ii) comparing different versions of the same operational ocean forecasting system and evaluating potential improvements and degradations before a new version is transitioned into operational status, (iii) comparing coarse-resolution “parent” and nested high-resolution “child” systems to quantify the added value of downscaling, and (iv) informing end users about the consistency and skill of the modeling products disseminated.

Developments in ocean modeling have clearly advanced to address the challenges associated with the increased resolution and its application to coastal areas, also responding to the high demand for providing 4D estimates of multiple oceanic variables at fine scales (Mourre et al., 2018; Fox-Kemper et al., 2019). Coastal modeling faces numerous challenges and issues such as downscaling and representation of open boundary conditions or land–sea and air–sea interactions (Kourafalou et al.et al., 2015a). Synergies between models and ocean observations are needed to face these challenges and improve representation of ocean processes (Kourafalou et al., 2015b; De Mey-Frémaux et al., 2019; Davidson et al., 2019). Additionally, it is worth mentioning the current lack of real-time and historical availability of observations in coastal areas, which limits the operational capability and reduces the potential of skill assessment operational services aiming to provide synthetic metrics addressing specific user needs (Révelard et al., 2021).

Within this context, HFR systems play a first-order role thanks to their unique ability to provide fine-resolution maps of the surface currents over broad coastal areas. This ability of HFR systems makes them particularly appropriate for the validation of numerical models in coastal areas, where other observations are scarce and/or their resolutions (i.e., in space or in time) are not high enough to capture the fine scale. Many HFR systems have therefore been used for this purpose in several regions of the Mediterranean Sea including the Northern Current area off Toulon (Berta et al., 2014a), the Ebro Delta area (Lorente et al., 2016, 2021; Ruiz et al., 2020; Aguiar et al., 2020; Sotillo et al., 2021), the northern Adriatic (Vilibić et al., 2016), the Gulf of Naples (Uttieri et al., 2011), the Ibiza Channel (Mourre et al., 2018; Aguiar et al., 2020; Révelard et al., 2021; Sotillo et al., 2021) and the Strait of Gibraltar (Lorente et al., 2019a; Aguiar et al., 2020).

An example of this added value of HFR data was recently shown in the multi-model comparison exercise performed in the Strait of Gibraltar in 2017 (Lorente et al., 2019a). In that case, the IBI-MFC model (Sotillo et al., 2015) was compared against their partially nested SAMPA (Sánchez-Garrido et al., 2013) high-resolution coastal forecast system to elucidate the accuracy of each system characterizing the Atlantic Jet (AJ) inflow dynamics. To this aim, HFR-derived hourly currents at the midpoint of the selected transect (square in Fig. 4a) were used as a benchmark. The scatter plot of HFR-derived hourly current speed versus direction (taking as a reference the north and positive angles clockwise) revealed interesting details (Fig. 4b): (i) the AJ flowed predominantly eastwards, forming an angle of 78∘ with respect to the north; (ii) the current velocity, on average, was 1 ms-1 and reached peaks of 2.5 ms-1. Speeds below 0.5 ms-1 were registered along the entire range of directions; (iii) westward currents, albeit in the minority, were also observed and tended to predominantly form an angle of 270∘ (i.e., towards the Atlantic), mostly related to intense easterly winds episodes (Garret, 1983; García-Lafuente et al., 2002; Menemenlis et al., 2007; Péliz et al., 2009; Reyes et al., 2015; Lorente et al., 2019a, b; Bolado-Penagos et al., 2021), as further detailed in Sect. 2.2.1. The scatter plot of SAMPA estimations presented a significant resemblance in terms of prevailing current velocity and direction (Fig. 4c). Although the time-averaged speed and angle were slightly smaller (0.9 ms-1) and greater (88∘), respectively, the main features of the AJ were qualitatively reproduced: maximum velocities (up to 2.5 ms-1) were associated with an eastward flow and an AJ orientation in the range of 50–80∘. Surface flow reversals to the west were also properly captured. By contrast, noticeable differences emerged in the scatter plot of regional IBI-MFC estimations (Fig. 4d): surface current velocities below 0.3 ms-1 were barely replicated, and the AJ inversion was only observed very occasionally. Despite the fact that IBI-MFC appeared to properly portray the mean characteristics of the eastward flow, the model tended to favor flow directions between 60 and 180∘ and to overestimate the current velocity, with averaged and maximum speeds around 1.17 and 2.80 ms-1, respectively.

In summary, HFR measurements are able to precisely assess the added value of the downscaling performed through the SAMPA coastal system with respect to the IBI-MFC regional solution in which SAMPA is nested. Overall, a steady improvement in the Atlantic Jet characterization is evidenced in model performance when zooming from regional to coastal configurations, highlighting the benefits of the downscaling approach adopted and also the potential relevance of a variety of factors at local scale, among others: a more refined horizontal resolution, a tailored bathymetry or the higher spatiotemporal resolution of the atmospheric forcing. Furthermore, SAMPA appeared to better reproduce the reversal events detected with HFR estimations, demonstrating the added value of imposing accurate meteorologically driven barotropic velocities in the open boundaries, imported from the NIVMAR storm surge model (Álvarez-Fanjul et al., 2001), in order to consider the remote effect of the atmospheric forcing over the entire Mediterranean basin, which was only partially included in IBI-MFC.

During the next phase of the Copernicus Marine Service, more focus will be on coastal downstream applications (e.g., very high-resolution ocean models integrated with coastal observatories) for a wide range of coastal stakeholders including ports and environmental agencies. Despite the significant progress in the field of coastal modeling, some storm-induced hazards are still not properly resolved (or even misrepresented) by ocean models due to a variety of factors (e.g., horizontal resolution that is too coarse, inadequate meteorological forcing, poor representation of land–sea interactions and the related river freshwater outflows, among others) as described by Sotillo et al. (2021). Within this framework, HFR might act as a monitoring cornerstone to calibrate and validate successive, upgraded versions of operational ocean forecasting models with the aim of better capturing extreme events in terms of strength, extension and timing (Lorente et al., 2021).

Aguiar et al. (2020) used the three HFR systems available in the western Mediterranean Sea (Strait of Gibraltar, Ibiza Channel – described in Tintoré et al., 2013, and Lana et al., 2016 – and Ebro Delta) to evaluate the impact of downscaling on the surface coastal circulation in the case of the Western Mediterranean OPerational forecasting system (WMOP) (Juza et al., 2016; Mourre et al., 2018). The authors showed that the time-average circulation in the coastal areas of the Ebro Delta and Ibiza Channel were improved through downscaling. In particular, the nested model showed a better representation of the small-scale coastal flow intensification at the mouth of the Ebro River and a refinement in the characterization of the circulation in the Ibiza Channel. Notice that HFR-Gibraltar, HFR-Ebro and HFR-Ibiza versus model comparisons are updated daily on the SOCIB WMOP web page https://socib.es/?seccion=modelling&facility=wmedvalidation (last access: 9 May 2022). Those HFR systems, among others, are also integrated in the IBISAR science-based data downstream service (Reyes et al., 2020a) – freely available under registration at https://www.ibisar.es/ (last access: 9 May 2022) – for visualizing, comparing and evaluating the performance of ocean current predictions in the Iberian–Biscay–Irish regional seas. IBISAR allows the identification of the most accurate ocean current dataset in a specific area and period of interest, thus facilitating decision-making for SAR operators and emergency responders. Lorente et al. (2022) consider the IBISAR service to be a successful example of the long-lasting engagement built in collaboration between HFR operators and end users (i.e., the Spanish Maritime Safety and Rescue Agency). Additionally, those HFR systems are also being used for IBI-MFC model assessment purposes by means of the NARVAL multiparameter and multi-platform validation tool (Lorente et al., 2019c) for IBI-MFC model validation.

Another added value of HFR systems is their use to improve model forecast through data assimilation (DA). DA aims at optimally combining observations and models to provide a better representation of the ocean dynamics. In this sense, HFR provides very valuable high-resolution observations in areas where satellite observations tend to suffer from limitations due to the vicinity of the coast (Vignudelli et al., 2019). While the assimilation of HFR measurements has been applied in many regions of the world since the first studies from Breivik (2001) and Oke et al. (2002), only a limited number of studies have been performed in the Mediterranean Sea. Marmain et al. (2014) assimilated radial velocity observations from the Toulon HFR system in a regional model in the Gulf of Lion. They showed how HFR observations can be successfully used to correct the wind forcing used to constrain the model coastal surface circulation. In the Ligurian Sea, Vandenbulcke et al. (2017) were able to correct surface currents and improve the representation of inertial oscillations after the assimilation of all the available hourly radial observations in a regional model of the area. Variational methods were also applied to improve model dynamics through multi-platform data assimilation including HFR in the southern Tyrrhenian Sea (Iermano et al., 2016) and in the Adriatic Sea (Janeković et al., 2020).

More recently, Hernández-Lasheras et al. (2021) specifically assessed the impact of assimilating HFR observations on the surface currents in the Ibiza Channel using the WMOP operational system. They compared the performance of both radial and total daily mean HFR-Ibiza surface currents (Tintoré et al., 2013, 2020) for correcting mesoscale and sub-mesoscale circulation using different initialization methods in an operational-like context. An independent Lagrangian validation performed by comparing non-assimilated and assimilated (without and with HFR) measurements with a set of 14 surface drifters (Tintoré et al., 2014) showed that the best results were obtained when using HFR total observations along with the traditional observation sources (i.e., satellite altimetry, sea surface temperature –SST, and Argo temperature and salinity profiles). After 48 h, the mean separation distance between virtual buoys and real drifters was reduced by 53 % compared to the simulation without any data assimilation and by 29 % compared with the simulation assimilating traditional observations only (as shown in Fig. 5).

To the best of the authors' knowledge, SOCIB WMOP (https://www.socib.es/?seccion=modelling&facility=forecast, last access: 9 May 2022) is presently the only system in the Mediterranean Sea including an assimilation scheme of HFR data in its operational chain.

Assimilation of HFR data into models is still computationally expensive and a complex issue, not to mention operational capabilities of such a procedure. Because of these constraints, the availability of real-time high-resolution HFR current fields has led to alternative solutions in order to obtain short-term prediction (STP) of surface coastal currents through the direct use of HFR historical and nowcast observations using different approaches (e.g., Zelenke 2005; Frolov et al., 2012; Barrick et al., 2012; Orfila et al., 2015; Vilibić et al, 2016).

The abovementioned studies develop and implement different STP approaches (harmonic analysis of the last hours, genetic algorithms, numerical models, etc.), which often require additional data or long training periods of data without gaps. Hardware failures due to power issues, communications or environmental conditions often result in spatiotemporal gaps within HFR datasets. Spatial gaps can be filled on a real-time basis, but the filling of long temporal gaps is not straightforward. Several gap-filling methodologies have been developed for HFR datasets: open modal analysis – OMA (implemented by Lekien et al., 2004, and further optimized by Kaplan and Lekien, 2007), data-interpolating EOFs – DINEOF (Beckers and Rixen, 2003; Alvera-Azcárate et al., 2005; Hernández-Carrasco et al., 2018b, Bourg and Molcard, 2021), self-organizing maps – SOMs (Kohonen, 1982, 2000, 2001; Hernández-Carrasco et al., 2018b), reduced-order optimal interpolation – ROII (Kaplan et al., 1997), optimal interpolation – OI (Kim et al., 2008), artificial neural network – ANN (Ren et al., 2018), variational analysis (Yaremchuk and Sentchev, 2011) and data-interpolating variational analysis in n dimensions – DIVAnd (Barth et al., 2021). HFR-derived short-term predictions were developed by Zelenke (2005), Frolov et al. (2012), Barrick et al. (2012), Orfila et al. (2015), Solabarrieta et al. (2016), Vilibić et al, (2016) and Abascal et al. (2017). More recently, Solabarrieta et al. (2021) developed a Lagrangian-based, empirical, real-time, short-term prediction (L-STP) system in order to provide short-term forecasts of up to 48 h of ocean currents from HFR data.

Through the NEURAL project (http://www.izor.hr/neural, last access: 9 May 2022), an innovative neural-network-based ocean forecasting system has been developed, providing gridded hourly surface current forecasts in the northernmost part of the Adriatic for the next 72 h. The forecasting system uses an unsupervised neural network algorithm, called self-organizing maps (SOMs, Kohonen, 1982; Liu et al., 2006), to train joint solutions coming from HFR measurements and a numerical weather prediction model as hourly surface currents and surface winds, respectively. Once the joint SOM solution has been trained, the surface current forecast follows the predicted surface winds that are the closest to the specific SOM solution (Fig. 6). Such a system requires a strong relationship between the predictor (i.e., surface winds) and the predictand (i.e., surface currents), which is largely found in coastal regions of the Mediterranean, yet it can be applied for any other combination of predictors and predictands. Also, high-frequency processes such as tides are removed from the system as being minor to the wind-driven dynamics, yet the tides can be added to the forecast.

The quoted northern Adriatic forecast system has been trained using 20 SOM solutions (so-called Best Matching Units, Liu et al., 2006) on HFR data measured between February and November 2008 conjoined with 3-hourly surface winds interpolated to 1 h resolution coming from the Aladin/HR operational model run once a day by the Croatian Meteorological and Hydrological Service (Tudor et al., 2013). The forecasting system performance was tested in the forecast (hindcast) mode during 2009 and 2010. Unfortunately, the HFR system has had substantial problems since 2010 and the antennas were eventually removed in the following years, resulting in a relatively short dataset that is possibly not sufficient for strong reliability of the forecasting system solutions. However, Vilibić et al. (2016) compared this SOM-based surface current forecast system (available online in http://jadran.izor.hr/neural/index.htm, last access: 9 May 2022) with the operational ROMS – Regional Ocean Modelling System (Shñhepetkin and McWilliams 2003, 2005) for the Adriatic, with the former lower showing biases and root mean square errors.

HFRs were used to investigate the upper ocean response to an extreme wind event in the Ligurian Sea (NW Mediterranean) during October–November 2018, as described in Berta et al. (2020). This work focused on the analysis of coastal sub-mesoscale structures, shaping surface currents and passive transport. The authors estimate the pattern and magnitude of kinematic properties (e.g., divergence and convergence as well as vorticity patterns) derived from surface currents measured by the HFR-TirLig network (Fig. 1) to characterize the evolution (before and after the event) of ocean scales at a few kilometers.

During the storm, sea surface vorticity (Fig. 7, top panels) and divergence (not shown but available in Berta et al., 2020) reach the order of the Coriolis parameter f, indicating ageostrophic activity typical of sub-mesoscale structures. The evolution of the sea surface structures suggested nonlinear interactions with the wind forcing. Considering the time series of wind speed and sea current properties (Fig. 7, lower panels) during and right after the storm (around 29 October), current magnitude increased approximately 4 times, while vorticity and divergence associated with the small features almost doubled. Such abrupt changes in horizontal currents and transport might also impact the vertical properties and in turn the ecosystem.

In the delta of the Ebro River (NW Mediterranean), the HFR-Ebro system observations have been crucial to capture the evolution of the most extreme Ebro River freshwater discharge event registered over the last 15 years in April 2018 (Ruiz et al., 2020). Results show the high impact of the freshwater pulse discharged on the surface circulation pattern, exhibiting a clear correspondence to high concentrations of satellite-derived chlorophyll a (Chl a hereinafter) concentration. Hovmöller diagrams of HFR-derived meridional and zonal currents indicate an increase in the southeastward velocity during the period of extreme river discharge. The proper representation of the basic oceanographic features of the HFR-Ebro, as the Ebro River impulsive-type freshwater discharged, was previously reported by Lorente et al. (2015).

This same region (i.e., Ebro Delta) was severely impacted by an exceptional storm in (19–24) January 2020, which surpassed the 99th percentile for several parameters (i.e., wind speed, significant wave height, wave period and surface current velocity) compared with the climatology and with a previous storm in January 2017. Particularly for this event, Lorente et al. (2021) have assessed the ability of the HFR-Ebro to characterize waves and currents under the record-breaking storm Gloria. By analyzing the data from the HFR-Ibiza and HFR-Gibraltar, the authors have also evidenced Gloria's remote effect in the Ibiza Channel and the Strait of Gibraltar, altering the usual water exchanges between adjacent sub-basins. Furthermore, the effect of Gloria was also manifested in the highest rates of particle dispersion at the Ebro River mouth on 21 January.

As in the previous case, the HFR system strategically installed in the Strait of Gibraltar (SoG) is considered an appropriate asset to effectively monitor the Atlantic Jet (AJ) inflow (Lorente et al., 2019b) and the water exchanges between the Atlantic Ocean and the Mediterranean Sea. The classical picture of the surface circulation is characterized by current pulses often exceeding 2 ms-1 and time-averaged northeastward speeds around 1 ms-1 in the narrowest section of the SoG (Fig. 4a). Complete collapse of the AJ and quasi-permanent inversion of the surface inflow during prevalent intense easterlies is a singular phenomenon that deserved detailed exploration (as previously mentioned in Sect. 2.1.2). Under this temporal premise, a monthly Hovmöller diagram was computed for HFR-derived zonal currents at the selected transect to easily detect a 2 d full reversal episode during March 2017, represented by black boxes in Fig. 8a. The event detected consisted of an abrupt interruption of the eastward inflow and complete reversal of the surface stream through the narrowest section of the SoG (Fig. 8b). The circulation in the easternmost region of the study domain was accelerated up to 0.8 ms-1, following clockwise rotation that likely fed the Western Alboran Gyre (WAG), which was out of the picture.

The prevailing atmospheric synoptic conditions were inferred from ECMWF predictions of sea level pressure and zonal wind at 10 m height (U10), as shown in Fig. 8c–d. A significant latitudinal gradient of sea level pressure was observed, with high pressures over the Gulf of Biscay and isobars closely spaced in the SoG, leading to extremely intense easterlies (above 10 ms-1) channeled through the strait due to its specific geometric configuration. Therefore, high pressures and intense, permanent and spatially uniform easterlies prevailed over the entire study domain, inducing westward outflow through the SoG as revealed by the 2 d averaged HFR circulation maps. Local wind forcing at this scale seemed to play a primary role in explaining such AJ collapse and the related inflow reversals, in agreement with previous studies (Garret, 1983; García-Lafuente et al., 2002; Menemenlis et al., 2007; Péliz et al., 2009; Reyes et al., 2015; Lorente et al., 2019b, Bolado-Penagos et al., 2021).

Tsunami early warning and alert represent an emerging and promising application of HFR. The main principle underlying the detection is that the abnormal surface current pattern induced by the orbital velocity of the tsunami wave train can be measured and interpreted in real time by an appropriate detection algorithm. The idea was first proposed by Barrick (1979), but it was only after the 2004 Indian Ocean disaster that the proof of concept was made on the basis of actual HFR data. It was shown numerically with simulated (e.g., Lipa et al., 2006; Gurgel et al., 2011) and real (e.g., Lipa et al., 2011, 2012; Dzvonkovskaya et al., 2012) events that the tsunami signature could be clearly seen in the HFR radial currents, and some appropriate detection algorithms were proposed. One strong point of HFR tsunami detection is that it is not bound to the nature of the source (i.e., seismic or atmospheric) and can be used as a useful complement to other warning systems in places where those are either not available or non-effective. Today, more than 20 real tsunamis have been detected “offline” with the reanalysis of HFR data. In view of the growing interest in these new capabilities of HFR, some radar manufacturers now provide commercial toolboxes along with their hardware system for the early detection of tsunamis; such systems have been installed in some places at risk (e.g., Vancouver Island, Canada; Oman; New Jersey, USA; Sagres in SW Portugal). To date, the only real-time detection was issued following a meteotsunami that occurred on 1 October 2016 in Tofino, BC, Canada (Dzvonkovskaya et al., 2017; Guérin et al., 2018).

However, no such HFR tsunami alert system has yet been installed in the Mediterranean Sea, even though there is a non-negligible tsunami hazard in this region, as witnessed by very destructive co-seismic events in recent history (e.g., Messina, Sicily, 1908). Some worst-case scenarios with a strong (M7.8) earthquake on the northern Algerian margin predict important tsunami waves with 3–4 m amplitude on the French–Italian Riviera (BRGM, 2007). Moderate earthquakes such as the M6.9 21 May 2003 Boumerdes–Zemmouri are sufficient to cause 1–3 m amplitude harbor oscillations within 40–60 min in the Balearic Islands, which would be the most impacted spot by seismic sources in northern Algeria (Wang et al., 2005; Sahal et al., 2009). The impact on the French Mediterranean coast impacted by a seismic source on the northern Algerian margin is shown in Fig. 9.

In addition to co-seismic tsunamis, frequent meteotsunamis (i.e., tsunamis of meteorological origin) have been reported in the Balearic Islands, named – “rissaga” (Jansa, 2007), Adriatic Sea – “šćiga” (Vilibić and Šepić 2009; Orlić, 2015), Sicily Channel – “marrobbio” (Candela et al., 1999; Zemunik et al., 2021), Malta – milgℏuba” (Drago, 2008), northern Persian Gulf (Kazeminezhad et al., 2021), Black Sea (Vilibić et al., 2021) and Aegean Sea (Papadopoulos, 1993). Even though these events have limited regional impact, they can cause severe local damage in harbors and bays due particularly to the micro-tidal regime, resulting in rapid sea level changes (Vilibić et al., 2021). Indeed, the strongest known meteotsunami in the Mediterranean Sea (and likely in the world) was in the Adriatic, the so-called Great Vela Luka Flood (June 1978), with an amplitude of 6 m and periods of about 20 min, as detailed by Vučetić et al. (2009), closely followed by the event from Ciutadella Harbor (Menorca island, Spain) in June 2006 (Jansa et al., 2007) with 4–5 m of amplitude. Meteotsunamis are caused by atmospheric disturbances combined with several further possible amplification mechanisms of the induced sea surface wave; these are mostly the so-called Proudman, Greenspan and shelf resonances in coastal areas, which can lead to strong harbor resonances in semi-closed basins (Orfila et al., 2011). Today, the generation of these meteotsunamis is better understood (Monserrat et al., 2006; Šepić et al., 2009, 2015; Vilibić and Šepić 2009; Ličer et al., 2017), but their prediction is still a very challenging task (Denamiel et al., 2019, Romero et al., 2019, Mourre et al., 2020).

When located in the areas affected by meteotsunamis, HFR-based tsunami early warning systems could be a useful complement to these forecasting systems, helping to issue specific alerts on the basis of the actual observed surface currents 20–40 km offshore a few minutes before the generation of extreme sea level oscillations. Note that tsunami early warning systems only require software updates of existing HFR and could be installed at reduced cost in some places. However, some strategic spots are not covered yet and would need a novel installation to monitor the travel directions of incoming waves from the most probable sources (northern Algerian earthquake, western Corsica submarine mass failure, northern Ligurian earthquake, etc.). Another related issue is extending the range of these HFRs, which would imply operating at lower-frequency bands (4–5 or 9 MHz) than those usually employed in the Mediterranean region (13, 16 or 25 MHz). An HFR prototype (i.e., Stradivarius radar) operating at 4.5 MHz with 200–300 km range was developed by Diginext Ltd. a few years ago for the Gulf of Lion as a proof of concept (Grilli et al., 2015). Such HFR systems can serve the double purpose of warning and characterizing abnormal surface current patterns arising from tsunami-like waves of seismic or atmospheric origin. As recently suggested by Domps et al. (2020), they can also be used as proxies for the observation of low-pressure fronts of atmospheric gravity waves that could lead to storm surge, if not meteotsunamis.

Coastal regions in the Mediterranean Sea are heavily inhabited, with strong tourism pressure and maritime activity resulting in human and industrial waste intentionally or accidentally dumped into coastal waters. These pollutants evolve according to their chemical transformation over time, are caught in the 3D general circulation, and are carried offshore and to other distant coastal areas by currents. For example, heavy metals or other chemical contaminants that may be present in semi-enclosed harbors (Tessier et al., 2011) could have important consequences for the ecosystem, and plastic litter is currently a massive and particularly harmful component (e.g., Ryan et al., 2009; Declerck et al., 2019) of the marine pollution in the Mediterranean Sea.

Furthermore, detecting, monitoring and cleaning up oil slicks following an offshore spill before they reach the coast is a major challenge. Consequently, monitoring, understanding and forecasting coastal dynamics are critical steps to develop adequate strategies to mitigate the effects of pollution in marine environments from or towards coastal areas. However, forecasting coastal dynamics is one of the most challenging issues in geosciences due to their strong space–time variability as well as the complexity of the processes controlling the dynamics that interact simultaneously over a broad range of time–space scales, as previously highlighted in Sect. 2.1.2. Thanks to the growing importance of HFR as a key element of coastal observing systems, coastal currents can nowadays be continuously measured in relatively large coastal areas, thus enabling the analysis of transport properties of the surface flow by means of different diagnostics based on the Lagrangian approach (i.e., addressing the effects of the spatiotemporal variability of the velocity field on drifted particles or tracers). Recent studies have demonstrated the potential of this land-based remote sensing technology for different applications in the field of tracking oil spills (Abascal et al., 2009), marine litter (Declerck et al., 2019) or phytoplankton (Hernández-Carrasco et al., 2018a).

In this regard, the trajectories of passive tracers are determined by the velocities measured by HFR (e.g., Ullman et al., 2006), often including the effect of subgrid-scale diffusion. When chemical pollutants are considered, additional processes should be included. For instance, when simulating oil spill trajectories, the advection term is a linear combination of the surface current velocity, the wind velocity and the stokes drift, and spreading, evaporation and emulsification should be included in the transport model.

Here we provide evidence to support the reliability of HFR currents for tracking substances in coastal areas. The Lagrangian validation has been performed using data from eight drifter trajectories available in the domain of the HFR area of coverage in the Ibiza Channel (HFR operated by SOCIB, Tintoré et al., 2013, 2020) during October 2012. We use the HFR velocity fields to compute the Lagrangian coherent structures (LCSs), which are very suitable to provide a template of the fluid flow transport (see Haller, 2015, and references therein), allowing the detection of transport barriers, which are of great relevance for marine dynamics. For example, LCS obtained from ridges of the finite-size Lyapunov exponents have been correlated with filaments of remotely sensed Chl a (Lehahn et al., 2007; Hernández-Carrasco et al., 2014, 2018a, 2020), sea bird foraging behavior (Tew Kai et al., 2009), the modeled extension of oxygen minimum zones (Bettencourt et al., 2015) and wind forcings (Berta et al., 2014b). At coastal scales, the dynamical picture in the Lagrangian frame has been analyzed using data from HFR currents to identify relevant small-scale transport barriers (Lekien et al., 2005; Gildor et al., 2009; Rubio et al., 2018), some of them focusing on coastal areas of the Mediterranean Sea (Haza et al., 2010; Berta et al., 2014b; Hernández-Carrasco et al., 2018a). Figure 10 shows the evolution of a set of virtual neutrally buoyant particles initially deployed on the northern (blue dots) and southern (black dots) flank of a given LCS measured from the HFR-Ibiza on 25 January at 17:00 UTC in 2013 (Fig. 10a). Although the location and magnitude of this LCS evolve in time, the LCS persists for several hours, manifesting the presence of a coherent transport barrier preventing the two sets of particles from being mixed up. A meridional LCS is formed and maintained during the simulated period, limiting water exchanges between the coast and the open ocean (Fig. 10a–c).

Another example of application of the Lagrangian properties derived from HFR focuses on the monitoring of the physical mechanisms that can influence the escape times of mesotrophic and/or polluted coastal waters to oligotrophic offshore areas, as described in Rubio et al. (2020). Using as input gap-filled HFR velocity fields a Lagrangian particle-tracking model provides the particle trajectories. From the Lagrangian model outputs, it is possible to infer the characteristic timescales for transport processes in the HFR footprint area by means of the escape rate of active particles (Fig. 11). Thus, HFR is shown to be an excellent tool to monitor conditions and identify the different scenarios that favor the local retention and dispersal of shelf waters in two study areas under the influence of ocean boundary currents.

Ocean eddies are ubiquitous, pervasive flow structures which dominate the ocean velocity field at several scales from the mesoscale to the local scale (Chelton et al., 2011). They play a fundamental role in sea dynamics, being responsible for the energy transfer among different scales (down to the dissipative range) as well as for their ability to transport nutrients, biomass, sediments and pollutants. Mesoscale eddies, produced by geostrophic instabilities, are not able to advance the energy transfer, being constrained by geostrophic and hydrostatic balance (Charney, 1971). When the balance is broken, the downscale may continue through inertia–gravity waves emitted from currents, ageostrophic instabilities and bottom boundary layer turbulence, which are responsible for the formation of sub-mesoscale eddies. At a lower scale, three-dimensional turbulence proceeds toward the dissipative range (McWilliams, 2019).

The presence of ocean eddies has become more evident in recent years thanks to the introduction of new oceanographic measurement techniques, while their exhaustive characterization would require synoptic time series of the velocity field in the ocean (Robinson, 1983). Such synoptic observations are made available only through satellite data. However, besides being limited to relatively large scales and preferably to the open ocean, they do not provide direct measurements of the total velocity field. Indeed, altimetry data can be used to retrieve the surface geostrophic field, which lacks a possibly important portion of the dynamics (Rinaldi et al., 2010; Conti et al., 2016). Other ways to observe eddies from satellites consist of the observation of their presence in the sea surface temperature or in the tracer field patterns, as displayed by ocean color (Robinson, 2010). Coastal HFRs overcome all the above limitations, enabling the detection and tracking of the time history of surface eddies down to sub-mesoscale at the cost of a reduced spatial extent.

Mandal et al. (2019) provide quite an extensive list of recent literature reporting sub-mesoscale features observed by HFRs, with examples of observations in various coastal areas from the Atlantic to the Pacific and Indian oceans (Shay et al., 1995, 2000; Kirincich, 2016b; Archer et al., 2017; Lai et al., 2017; Arunraj et al., 2018). It is worth noticing that such features may have a strong vertical signature that HFR data fail to account for and therefore need to be complemented by further information spanning from direct measurements of the horizontal and vertical velocity profile to indications indirectly derived from, e.g., satellite turbidity measurements (see discussion in Uttieri et al., 2011).

A vast HFR network now covers the northwestern Mediterranean coastal areas (see Lorente et al., 2022), which is characterized by significant mesoscale variability and eddy generation that, in some cases, shows recursive and seasonal patterns. Allou et al. (2010) used HFR to observe and characterize vortex structures, mostly anticyclonic, in the Gulf of Lion. They also argued they were correlated with specific wind patterns. Schaeffer et al. (2011) improved the analysis and employed both HFR measurements and numerical modeling to analyze the eddy-generating mechanism. They found it is primarily influenced by wind forcing and its interaction with topographic constraints (northerly offshore wind), as well as freshwater input from the Rhone river (southerly onshore wind). The combination of HFR, in situ observations and modeling tools allowed Guihou et al. (2013) to identify an anticyclonic coastal eddy which was generated in front of Nice by a meander of the Northern Current and advected downstream toward the Toulon area, interacting with the mean circulation. More recently, the analysis of the long 2012–2019 HFR time series in Toulon allowed identification of cyclonic and anticyclonic recurrent eddies mainly generated by wind and boundary current undulations (Bourg and Molcard, 2021).

The development of monitoring networks providing long time series of data is making automatic eddy detection methods more and more topical and important. Generally speaking, existing eddy detection algorithms can be divided into three families: (i) those that are based on the geometrical features of the velocity field, typically in terms of streamline closedness, winding angle or vector geometry (Sadarjoen et al. 1998; Heiberg et al., 2003; Ebling and Scheuermann 2003; Nencioli et al., 2010); (ii) those based on dynamical characteristics, such as parameters quantifying the eddy intensity and its vorticity (Jeong and Hussain 1995; Fang and Morrow 2003; Isern-Fontanet et al., 2003; Morrow et al., 2004); (iii) and hybrid methods based on the combination of geometric and dynamical criteria (Mkhinini et al., 2014; Conti et al., 2016, see also the extensive review in the paper by Le Vu et al., 2018).

Algorithms specifically devised for HFR data are very few. The methods tested in the Mediterranean Sea (Caldeira et al., 2012) are limited to those by Nencioli et al. (2010) and Bagaglini et al. (2020). The former, even though developed for HFR data (and for high-resolution numerical model outputs), has found a widespread range of applications to observations collected by different platforms, as witnessed by current oceanographic literature (Liu et al., 2012; Dong et al., 2014). It is a method based on the geometry of the velocity vectors. It was conceived for geostrophic or quasi-geostrophic recirculating features showing very little divergence. For this reason, it is very suitable to describe mesoscale eddies but may fail in detecting sub-mesoscale ones, which often are characterized by divergence or convergence and by a high degree of deformation of the velocity field geometry. YADA (Yet Another eddy Detection Algorithm), developed by Bagaglini et al. (2020), was conceived specifically to overcome this limitation and be utilized to automatically detect sub-mesoscale eddies, which may exhibit highly non-geostrophic characteristics. It is a hybrid method, which focuses on both the dynamical and geometric features of the velocity field, first identifying the local extrema of a dynamical field characterizing recirculations (e.g., the local normalized angular momentum, see Mkhinini et al., 2014; the Okubo–Weiss parameter, Okubo, 1970; Weiss, 1991), similarly to the first step from AMEDA (Angular Momentum Eddy Detection and tracking Algorithm) defined by Le Vu et al. (2017), and thereafter analyzing the streamline geometry in a neighborhood of the extremum. YADA (Bagaglini et al., 2020) has been successfully applied to 1 km resolution HFR data from the Gulf of Naples, showing its ability to identify strongly asymmetric, convergent or divergent sub-mesoscale eddies. Its application to coastal HFR data from other areas of the western Mediterranean is presently under way.

The HFR system from the LaMMA Consortium (described in Lorente et al., 2022) covers part of the Ligurian Sea and the Tuscany Archipelago, which is a shallow area separating the Ligurian and Tyrrhenian basins, bordering the Corsica Channel eastward, with complex topography and coastal morphology, also due to the presence of several islands (Elba, Capraia, Montecristo, Gorgona). Sea dynamics are strongly influenced by seasonality and characterized by the presence of the Tyrrhenian boundary current and its bifurcation, the Eastern Corsica Current (Astraldi and Gasparini, 1992; Millot, 1999; Vignudelli et al., 2000). Through drifters, in situ data and a numerical model, Poulain et al. (2020) studied the area in the summer season. They found prevailing southward current flowing next to the Italian coast, then turning westward and northward, encountering Elba island. A further eastward motion led to the formation of an anticyclone centered on Capraia island, which exhibited variations correlated with wind forcing. The presence of an anticyclone north of the Corsica Channel in summer and autumn seasons was previously documented by Ciuffardi et al. (2016) by means of in situ profiles and altimetric data. Furthermore, they argued that the characteristics of the anticyclone (position and size) may affect the general circulation by isolating the Tyrrhenian and Ligurian basins, mostly in summer. These hydrodynamics features can influence the concentration of floating marine litter, which was shown to be particularly high in certain periods of the year (Fossi et al., 2017).

Here, we show the application of both YADA and the Nencioli et al. (2010) algorithms to the surface currents derived from the HFR-LaMMA system during the year 2019.

Figure 12 reports four eddies detected by YADA and the corresponding surface currents in August 2019.

Figure 13 shows the results of the application of the algorithm by Nencioli et al. (2010) to the whole year 2019, which provided a seasonal census of anticyclonic and cyclonic eddies in the area sampled by the HFR-LaMMA system. The HFR coverage was not uniform during the year, with generally lower percentages for the warmer seasons (i.e., spring, summer). The area north of Elba island was characterized by the highest eddy activity throughout the year, with a predominance of anticyclonic eddies in colder seasons and a clustered pattern in summer, showing anticyclonic eddies east of Capraia island and cyclonic ones toward the coast.

The number of detected eddies depended also on the availability of HFR data; indeed, a smaller number of eddies was found in spring (28) with respect to the other months (62 winter, 68 summer, 62 autumn). Median eddy life span was around 1.125 d for all seasons except in spring, although some were able to survive up to 6 d.

Although the present work is preliminary, it may lay the foundation for a detailed analysis concerning the seasonal features of eddy activity within the Tuscany Archipelago and its effect on the general circulation. Surface currents from HFR can be combined with numerical model outputs to bridge the gap relative to the spatial and temporal coverage of data and to improve the representation and forecast of the real sea state: HFR observations enhance numerical simulations by resolving fine-scale processes in intricate regions with complex geometry configurations. In turn, hydrodynamic models can reciprocally serve as integrative connectors of sparse in situ observations and gappy HFR surface current maps by offering a seamless predictive picture of the three-dimensional ocean state.

The necessity to preserve the marine ecosystem equilibrium and the water quality has fostered the use of HFR data in supporting coastal zone management and assessing the variability in the dynamics of marine ecosystems. In particular, HFR data have been used worldwide to address ecological and water quality issues such as (i) to understand the transport and retention processes of plankton or wastewater discharge in some regions of NW Spain (Ria de Vigo) (Piedracoba et al., 2016) and the western Mediterranean (Hernández-Carrasco et al, 2018a), at coastal upwelling fronts off central California (Bjorksted and Roughgarden, 1997) and in Monterey Bay (Coulliette et al., 2007); (ii) to investigate the enhancement of productivity due to the retention of phytoplankton within the flow in the Santa Barbara Channel (Brzezinski and Washburn, 2011); and (iii) to analyze the relationship between populations of larval and juvenile fishes and the mesoscale flow field in the California Current System (Nishimoto and Washburn, 2002).

Here we present examples of application of HFR data to investigate ecological questions in three Mediterranean coastal areas: the Gulf of Naples (GoN, Tyrrhenian Sea), the Gulf of Manfredonia (GoM, western Adriatic Sea) and the Malta–Sicily Channel (mid-Mediterranean Sea).

In the GoN, continuous observations of the HFR current fields highlighted several characteristics of the surface circulation and water exchange between the interior of the gulf and the neighboring open Tyrrhenian Sea (Cianelli et al., 2013). An oscillating plankton population dynamic has also been frequently observed in the GoN, at Long-Term Ecological Research station MareChiara (LTER-MC), where plankton abundance has been monitored weekly since 1984. A proof-of-concept study (Cianelli et al., 2017) was thus conducted in order to characterize the spatial scales and the provenance of phytoplankton assemblages detected at LTER-MC and to dissect processes regulating plankton dynamics.

The study focused on a year-long analysis carried out for 2009, which was characterized by a very accurate estimate of the surface dynamics, with a reduced number of gaps among ecological measurements and HFR data. The approach followed these conceptual steps: (i) reconstruction of the annual and seasonal regimes of HFR currents detected at the LTER-MC site and (ii) computation of Lagrangian backtracking simulations advecting virtual phytoplankton patches (VPPs) in the HFR field (Fig. 14a). VPPs were released at the LTER-MC site on the dates of the weekly oceanographic campaigns and tracked backward, thus allowing the estimation of the positions of the VPPs up to 4 d (i.e., 96 h) prior to arrival at LTER-MC. The next step was (iii) identification of the prevailing directions from which the VPPs arrive at the LTER-MC site as resulting from backtracking simulations (Fig. 14b), also allowing the definition of the spatial distribution of the VPP origin zones (not shown) in the GoN. This was followed by (iv) comparison among backtracked Lagrangian reconstruction and ecological analysis based on salinity and Chl a data obtained through weekly sampling at LTER-MC (Fig. 14c) as well as (v) identification of different modes of coupled physical and ecological functioning in the GoN as resulting from physical transport and biological processes (“allogenic” and “autogenic” factors, respectively).

The results showed an alternation in plankton dynamics between phases reflecting the influence of coastal (green) and offshore (blue) circulation patterns on the biological community. The phytoplankton community detected at LTER-MC generally originated from the coast, whereas the offshore inflow marginally changed the main traits of phytoplankton assemblages. Backtracking simulations and biological data strictly agree, highlighting that the plankton community at LTER-MC during 2009 was affected by the alternation of coastal and offshore influence.

Biological autogenic factors drive the modifications of coastal phytoplankton communities during coastal “green” phases, thus suggesting that the GoN tends to retain the same communities via coastward circulation, especially during summer. Physical allogenic factors are important in driving dilution and species advection of coastal phytoplankton during the offshore “blue” phase. This marked alternation between coastal and offshore water masses acts to promote phytoplankton diversity because the dilution in the phytoplankton density may decrease the impact of the dominating species over the available resources.

The integration of long-term biological data and high-resolution current fields represents an optimal tool to investigate the role of surface circulation in structuring the marine plankton community, thus confirming the value of HFR systems to analyze the seasonal fluctuations in marine ecosystems dynamics and to unveil the mechanisms of coastal connectivity.

The GoM is a well-known recruitment area in the Adriatic Sea (Sciascia et al., 2018; Corgnati et al., 2019a). In this region, HFRs have been used to understand the role of ocean currents in the recruitment of small pelagic fishes (i.e., European sardines, Sardina pilchardus). Figure 15 shows residence times within the GoM, estimated using trajectories of virtual drifters computed from the surface currents measured by HFRs. Months with high (October) and low (February) residence times are associated with weaker and stronger surface currents in the central area of the gulf. The relatively short (<12 d) average residence times have shown that local spawning is less likely to take place than transport to the gulf from remote spawning areas through advection pathways. Results agree with otolith measures, suggesting that the arrival of larvae within the gulf is characterized by repeated pulses from remote spawning areas that are likely to play a fundamental role in maintaining the nursery.

The Malta–Sicily Channel is both the most fished area in western Mediterranean Sea and a very important hot spot of biodiversity (Médail and Quézel, 1999). In this region, Capodici et al. (2018) combined the HFR-CALYPSO surface currents together with satellite images of Chl a and SST to explain physical driving mechanisms that can help interpret ocean productivity and plan maritime activities in a more adaptive and proactive way. On the one hand, as mentioned by the authors, monitoring water quality data provides valuable insight into the understanding of processes driving spatial and temporal changes in productivity at sea (Behrenfeld et al., 2006). Chl a and SST are generally accepted as proxies for water quality and are very helpful to detect upwelling events, which frequently occur along the Sicilian coast. In particular, sea currents are responsible for dispersion, transport or retention of nutrients among which Chl a is widely used as a proxy variable; moreover, current jets or eddies are often observable as cold or warm areas in SST maps, respectively. Even if both Chl a and SST maps are usually retrieved by means of satellite data maps, cloudiness often reduces the satellite data availability; thus, temporal aggregated products (e.g., at 8 or 16 d) are the only data available. In this framework, the integration of sea surface current data provided by HFRs can fill the gap of knowledge due to the inadequate temporal (and sometimes spatial) resolution of these water quality maps. Capodici et al. (2018) used principal component analysis – PCA (Preisendorfer, 1988) to firstly extract the dominant spatial patterns of these variables in 2013 (Fig. 16, left panels) and to secondly quantify the degree of correlation between SST (or Chl a) and HFR-derived surface current spatiotemporal patterns (see Fig. 16, bottom right panel). The spatial correlation analysis suggests the importance of current advection in phytoplankton transport, being characterized by fringes where very high positive correlation areas are surrounded by very high negative ones and vice versa.

Moreover, the spatial distributions of time-averaged radar currents and corresponding TKE (total kinetic energy), EKE (eddy kinetic energy), and the absolute value of the products between temporal fluctuations of the deviations from the time-averaged zonal and meridional velocity components (ReS) have shown swirling areas for the first time at high spatiotemporal resolution. The trapping zones, characterized by low values of TKE (see Fig. 16, top right panel), EKE and ReS at their cores, may help explain why this channel is particularly rich in pelagic species, highlighting how continuous high-resolution HFR-derived surface currents can improve decision-making in fisheries resource management.