An abrupt mixing event in the upper ocean is investigated in the Northwestern Mediterranean Sea using gliders, a new ocean monitoring technology, combined with regional atmospheric model outputs and mooring data. Intense winds (up to 20 m s−1) and buoyancy forcing during December 2009 induced strong vertical mixing of the upper ocean layer in the Balearic Sea. High-resolution data from a coastal glider reveal a surface cooling of near 2 °C and the deepening of the Mixed Layer Depth (MLD) by more than 40 meters in the center of the basin. Comparisons between glider and ship-emulated sections of hydrographic profiles show that the glider data make visible the small-scale spatial variability of the MLD. The heat content released to the atmosphere by the upper ocean during this mixing event exceeds 1000 W m−2. A simulation from the Weather Research and Forecasting model reports values consistent with these observations. Additionally the atmospheric numerical simulation shows the development and evolution of a cyclone located south of the Balearic Islands. This cyclone is likely to be responsible for the wind intensification and the consequent air-sea energy exchanges that occurred in the study area during this period.
 Characterization of the spatial and temporal variability of the Mixed Layer Depth (MLD) is essential to developing a better understanding of the exchanges of air-sea fluxes (wind stress, heat and fresh water) that occur daily at the ocean surface. The MLD perturbs the sea surface temperature, which has direct implications for atmosphere and ocean heat budgets [Roads et al., 2003]. Different criteria to estimate the MLD have been studied with successful results [de Boyer Montégut et al., 2004; Lorbacher et al., 2006; Chu and Fan, 2010], and de Boyer Montégu et al.  developed a climatology (2° resolution) of the MLD for the global ocean. In the Mediterranean Sea, D'Ortenzio et al. produced a climatology (0.5° resolution) that reproduces the MLD associated with the main features of the surface circulation and deep-water formation areas.
 For shorter space and time scales the MLD is less well studied, as this requires in-situ experiments under specific (often extreme) weather conditions to characterize the high resolution horizontal and vertical extension of the MLD associated with atmosphere-ocean mesoscale processes. For example,Weller (1991)carried out an exhaustive study on air-sea interactions in the Atlantic Ocean combining mooring, drifters, SeaSoar and ADCP data. In the Northwestern Mediterranean Sea the exchanges of energy between mesoscale atmospheric fronts and the upper ocean are especially significant during the winter period when northwesterly winds are dominant [Jansá, 1987; Flamant, 2003], however studies to investigate the MLD under the influence of strong winds in this area are scarce and based on mooring data [Marty et al., 2002], upward looking ADCPs [Schott and Leaman, 1991] or coastal weather stations, model outputs and ship CTD observations [Mertens and Schott, 1998]. The hydrographic observations from these previous studies have usually been collected by performing CTD casts around the moorings or by defining transects with large spatial sampling intervals (5 to 10 km). In this context, the objective of this study is to investigate the upper ocean response to intense winds and buoyancy forcing in the Western Mediterranean using high resolution glider observations, in situ buoys and an atmospheric numerical simulation. A major advantage of using gliders is the acquisition of high spatial resolution hydrographic data (0.5 km) in wintertime, when CTD sampling from ship surveys is complicated by frequent adverse weather conditions. First we characterize the MLD before and after the mixing event, estimating the heat fluxes exchanges from the glider data. These results are compared with those obtained from ship-emulated sections of hydrographic profiles to assess the impact of using the higher resolution glider data. Then, an atmospheric numerical simulation is considered, to show the heat flux exchange and the potential mechanism responsible for the ocean mixing event. Finally, a discussion of the results and concluding remarks are given.
2. In-Situ Data and Atmospheric Model
2.1. Glider Data
 In the framework of the IMEDEA observational program for autonomous underwater vehicles in the Balearic Sea (Western Mediterranean), a glider mission was carried out between 9 to 20 December 2009 (Figure 1a). As in previous missions (see details in work by Ruiz et al.  and Bouffard et al. ), the data acquisition strategy consisted of performing a section almost perpendicular to the Balearic and Northern Currents, that flow along the north coast of Mallorca and along the Iberian Peninsula respectively [La Violette et al., 1990]. 700 CTD profiles were collected, between surface and 180 m, covering a total track length (go and return) of about 270 km. The glider data processing included the thermal lag correction for the un-pumped CTD unit, standard on Slocum gliders [Garau et al., 2011]. Final profiles have a 0.5 km horizontal along-track resolution and were averaged in the vertical to 1 db.
 To compute the MLD different approaches are available [Kara et al., 2000; Lorbacher et al., 2006]. Here, the MLD that is commonly defined as the layer between surface and the depth where hydrographic measurements (temperature, salinity and density) are quasi-constant, is computed followingde Boyer Montégut et al. . The specific criteria considered for the computation is T(10 m) - T(h) = 0.2 °C, where T(h) is temperature at depth h and T(10 m) is temperature at 10 m depth, as the first data available from the glider for all the profiles is at 10 m depth.
 To assess the impact of high-resolution glider data on the estimation of the MLD, a second dataset at lower resolution has been produced; through sub-sampling the glider data at 5 km horizontal resolution and giving rise to a dataset we have called “ship-emulated” data.
 Averaged hourly data from the network of buoys operated by Puertos de Estado (Spanish Port Authority) are used for the study at 4 locations (Figure 1a). These buoys provide information on atmospheric and oceanographic parameters, near Mallorca (Mallorca buoy) where the glider mission started and finished, and also in the vicinity of the end point reached by the glider near the mainland (Tarragona buoy). Two additional buoys, located to the north (Begur buoy) and south (Valencia buoy) of the glider mission, are also used. The specific variables analyzed from the buoys are wind speed and direction, sea level atmospheric pressure, sea surface height and air temperature at 2 m.
2.3. Atmospheric Model
 The Weather Research and Forecasting (WRF) Model (v3.00 [Skamarock and Klemp, 2007] was implemented in a nested 2 grid configuration (Figure 1b): the largest domain, with a horizontal resolution of 30 km, approximately covers the western Mediterranean, the inner domain, with a horizontal resolution of 7.5 km, covers the area 35.4 N–43.2 N, 1.9W–7.6E, which includes the Balearic Islands. The coarser grid reproduces the large-scale features, which force the local dynamics in grid 2 at each time step. The simulation is performed using a two-way nesting technique, starting at 00:00 UTC 1stNovember 2009 and lasting 2 months. Forty-six vertical levels are employed, more closely distributed in the upper levels. Initial and boundary conditions are provided by the NCEP FNL analysis (http://dss.ucar.edu/datasets/ds083.2). A full set of parameterization schemes are included in the model for microphysics, convection, turbulence, soil processes, boundary layer processes, and radiation. In the model configuration used here, the following parameterization schemes have been selected: the WRF Single-Moment 5-class scheme microphysics [Skamarock et al., 2008], the Kain cumulus parameterization on the coarser grid [Skamarock et al., 2008], the Rapid Radiative Transfer Model (RRTM) for long-wave radiation (based on work byMlawer et al. ), the Dudhia  scheme for shortwave radiation, and the Noah Land Surface Model [Skamarock et al., 2008]. The planet boundary (PBL) scheme used is the Mellor-Yamada-Janjic scheme [Janjic, 2002]. SST forcing was derived from a high resolution (∼5 km) daily SST product [Marullo et al., 2007]. Fluxes data from WRF have been interpolated to the exact glider positions using cubic splines.
3.1. Oceanographic Conditions
 During the glider mission, relatively cold (15 °C) and salty (>38 PSU) Mediterranean waters occupied the upper 100 m near the Spanish mainland coast, while on the island side, relatively warm (17 °C) and fresh (<38 PSU) of recent Atlantic origin was found in a shallower layer of about 60 m (Figures 2a and 2b). Dynamics of the area is dominated by two main currents [La Violette et al., 1990], that have been intensively studied using gliders and altimetry in the last 5 years [Bouffard et al., 2010] and that can be briefly described as the Northern Current flowing southwestward along the Spanish coast and the Balearic Current that progress northeastward along the Balearic coast. Intense mesoscale variability associated to this general pattern has been also reported [Tintoré et al., 1990].
 Density fields and MLD are shown in Figures 2c and 2d. For the ‘go’ transect the MLD is found at 40 m in the center of the basin, which is in good agreement with the low resolution (1 degree) climatology values [de Boyer Montégut et al., 2004]. At the ends of the transect, associated with the along-coast currents, the MLD is located at deeper levels (about 90 m). The climatology does not have sufficient resolution to detect this spatial variability at the edges of the domain (not shown). The stability of the water column has been also illustrated through computing the Brunt-Väisälä frequency (Figures 2e and 2f), which confirms that maximum values are associated with the pycnocline, which is shallower in the center of the basin. The equilibrium of this upper ocean layer, as sampled by the glider during the first 4 days, was abruptly modified during the second part of the mission (glider ‘return’ transect from mainland to Mallorca). On 13th December, atmospheric pressure started to drop off drastically, decreasing from 1025 hPa to values near 995 hPa by 15th December (Figure 3a). In this short period, wind velocity increased from less than 5 m s−1 to maximum values of 20 m s−1, inducing significant wave heights of up to 6 m (Figure 3b). Under these adverse sea conditions, gathering hydrographic data by ship would be a difficult and risky task. However, the autonomous underwater platform was able to successfully accomplish the transit from the mainland to Mallorca.
 The abrupt change in atmospheric conditions induces an immediate response in the upper ocean. Potential temperature at the surface ocean diminished from 16 °C to about 14.5 °C and the momentum and buoyancy forcing causes a sinking of the MLD. While the pycnocline is maintained at almost the same depth at the boundaries of the study area, it sinks by about 40 meters in the middle of the basin. A closer inspection of the thermohaline structure reveals additional changes in the southern part of the domain, near the north of Mallorca coast. In this area the structure located between the surface and about 70 m, characterized by relative warm and fresh water (typical of the Balearic Current, Pinot et al. ), was split in two parts; a main core near the coast and a small-detached filament centered at about 40°N (Figure 2b). After the mixing event, the maximum value of N2 at the pycnocline is still of the order 50 (radians s−1)2 but now it is located at 80 m depth at both the ends and at the center of the basin (Figure 2f). The heat content of the upper ocean layer has been estimated for the glider data (go and return tracks) and the difference reveals loss of about 1100 W m−2 in the center of the basin (Figure 3c).
 A first look to the glider and ship-emulated sections of hydrographic profiles indicates that, as expected, the large-scale pattern in both samplings is very similar (Figures 2a and 2b and auxiliary material). However, in the glider data it is possible to determine the small-scale spatial variability of the MLD, which has an impact on the estimation of the upper ocean heat content. The smoothed MLD estimated from the ship-emulated dataset in the south part of the domain, between latitude 39.85 N and 40.1 N, shows a substantial underestimation (about 20%) of the heat content (Figure 3c) when compared with the value estimated from the glider data.
3.2. Local Forcing and Atmospheric Model Outputs
 The 4 buoys, located in the Northwestern Mediterranean area, gathered the atmospheric conditions for the whole period of the glider mission. Comparisons between the wind speed and pressure from the buoys and model simulation show good agreement (auxiliary material), demonstrating that the chosen model parameterization accurately reproduces the local atmospheric variability and allows us to use the regional atmospheric simulation to study the air-sea interaction mechanisms forcing the glider observed ocean mixing event.
 Heat fluxes (sensible and latent) from the WRF model reveal a significant loss of energy from the ocean on 14thDecember. Latent heat fluxes reach maximum values of about -750 W m−2while the sensible heat released to the atmosphere is of the order of -400 W m−2 (Figures 4a and 4b). The heat balance given by the WRF simulation is about -1200 W m−2, which is in good agreement with the estimation from the in-situ glider observations (seeFigure 3c).
 The WRF output also allows us to understand the important role of the latent heat during this event. From 13th December onwards, the atmospheric simulation reveals the arrival of cold air and intense northwesterly winds. The atmospheric pressure from the model shows a core of low pressure (around 990 hPa) located south of the Balearic Islands at the same time that the glider was performing the ‘return’ sampling towards Mallorca (14th December). Intense winds of up to 25 m s−1 (Figure 4c) crossed the study area cooling the surface layer and producing the mixing of the upper ocean layer, as described above, and a net surface heat loss across the Balearic Sea.
 In addition, the potential origin of the atmospheric forcing has been investigated using the WRF model. Mean sea level pressure and RMS for the 14-16 December period is shown inFigure 4d. The core of the cyclone is detected near the Moroccan coast by November and it then advanced northeast towards the south of the Balearic Islands, overlapping with the period of glider sampling. By the end of December the storm had progressed southeast, towards the African coast, where it finally dissipated (Figure 4d).
4. Discussion and Concluding Remarks
 We have characterized an air-sea interaction in the Western Mediterranean using new technologies such as gliders; regional atmospheric model outputs and data from buoys are also used. From 9th to 13thDecember 2009, the high-resolution hydrographic data reported the characteristic pattern of the thermohaline structure in the Balearic Sea. The situation changed drastically from 13thDecember onwards. Simultaneous to intense winds and buoyancy forcing a glider recorded, at high spatial resolution, the strong vertical mixing event of the upper ocean layer. Surface ocean water suffered a sudden cooling of near 2 °C and the MLD deepened by more than 40 m in the center of the basin. Comparisons between glider and ship-emulated sampling demonstrate some of the advantages of using these autonomous platforms. Gliders can provide a more detailed picture of the spatial variability of the MLD depth and therefore a more accurate estimation of the ocean heat content is possible. However, gliders also have some limitations, such as the low speed of displacement [Rudnick and Cole, 2011], which affect the synopticity of the data. In our study, as the event lasted several days, the 4 day glider transect is viewed as sufficiently synoptic to accurately represent the significant changes in MLD that occurred during this period.
 The amount of heat content released to the atmosphere by the upper ocean during the mixing event was estimated from the glider data and found to exceed 1000 W m−2. The simulation of heat fluxes from the WRF model reports values consistent with these observations. It is also worth noting that similar values have been reported in the area under equivalent wind conditions [Mertens and Schott, 1998; Schott and Leaman, 1991]. The slight differences between glider and model fluxes (less than 100 W m−2) could be explained by the lateral advection (which should be estimated from accurate ADCP velocities not available for this study) and errors in the SST forcing used by the WRF model [Marullo et al., 2007]. A regional atmospheric model simulation carried out for the glider sampling period reveals the development and evolution of an atmospheric cyclone located south of the Balearic Islands. This cyclone appears to be responsible for the intensification of winds from the north and the consequent increase in air-sea exchanges in the basin.Campins et al.  and Lebeaupin Brossier and Drobinski  have reported similar wintertime events, characterized by intense, cold and dry winds from Northern Europe.
 This study demonstrates that autonomous underwater platforms represent a valuable new tool for investigating the air-sea interactions in extreme conditions, as they can be operated remotely and collect data at a high horizontal resolution (0.5 km). Moreover, glider data could contribute to validate atmospheric model products such as heat fluxes. In the coming years we plan to maintain and increase our glider activities in the Western Mediterranean and to learn more about the capabilities and benefits of these platforms in studying the upper ocean.
 This work has been partially supported by Gliderbal and MyOcean projects, funded by the Govern Balear (AAEE0051/08) and the European Commission respectively. We would like to acknowledge the support of the technicians from the TMOOS Department at IMEDEA (CSIC-UIB), for their collaboration during the glider preparation, launch and recovery, and of Puertos del Estado in providing the mooring data. Emma Heslop and two anonymous reviewers provided valuable comments that helped improving the manuscript. The FNL data (http://dss.ucar.edu/ds083) are from the Research Data Archive (RDA) maintained by the Computational and Information Systems Laboratory (CISL) at the National Center for Atmospheric Research (NCAR). NCAR is sponsored by the National Science Foundation (NSF).
 The Editor thanks two anonymous reviewers for their assistance in evaluating this paper.