Temperature and salinity profiles from different sources are collected, which are measured using CTD (conductivity–temperature–density profiler), PFL (autonomous profiling floats including ARGO floats), XBT (expendable bathythermograph), and MBT (mechanical bathythermograph). The World Ocean Database (https://www.nodc.noaa.gov/OC5/SELECT/dbsearch/dbsearch.html, last access: 1 February 2017) is the main source. Apart from this, data from the CORIOLIS data center (http://www.coriolis.eu.org/Data-Products/Data-Delivery/Data-selection, last access: 1 February 2017) and several cruises conducted by individual institutions are also used in this analysis. The bathythermograph profiles were depth corrected based on Cheng et al. (2014). A total of 13 891 temperature profiles were made for the Red Sea (approximately 14 % of these profiles have salinity measurements) from 1934 to 2017.

These profiles are quality checked according to the procedure given in Boyer and Levitus (1994). In the duplicate check, all the profiles within a 1 km radius and taken on the same day are considered duplicates and are removed from the main dataset. The levels in the profile with large inversions in temperature (inversion ≥ 0.3 ∘C) are flagged and removed. If three or more inversions are present, then the entire profile is removed. The levels with extreme gradients ≥ 0.7 ∘C are also removed from the profile. Since the present work is more focused on the changes in the upper layer of the ocean (from the surface to a 150 m depth), profiles with low resolutions in the upper layers are removed. Almost 50 % of the profiles have resolutions of < 5 m, while 7 % of the profiles have poor resolutions (resolutions of > 25 m).

Out of the total 13 891 profiles analyzed, 11 212 profiles passed the quality check from CTD (690), PFL (1385), XBT (5507), and MBT (3630), and the spread is shown in Fig. 1. More than 80 % of these profiles are positioned along the middle of the Red Sea, with a sufficient number of profiles for each month (Fig. S1 in the Supplement). The yearly and monthly distributions of the temperature profiles lie along the middle of the Red Sea and are given in the Supplement (Figs. S2–S3). As part of the quality check, 2679 profiles were removed from the main dataset. A total of 2063 salinity profiles are available for the entire Red Sea (Fig. S4 in the Supplement). MLD is estimated based on the temperature profiles due to the increased number and sufficient monthly coverage compared to that of salinity. The distribution of the temperature profiles used in this analysis is shown in Fig. 1.

The monthly mean values of heat fluxes and wind stress data are provided by TropFlux at a 1∘ × 1∘ spatial resolution for the period 1979–2016, which are used to check the influence on MLD variability (http://www.incois.gov.in/tropflux_datasets/data/monthly/, last access: 1 February 2017). TropFlux captures better variability and less bias than the other available fluxes and wind stress products (Praveen Kumar et al., 2012, 2013). Since evaporation is not provided by TropFlux, the monthly mean values of evaporation from OAFlux (from 1979 to 2016 and 1∘ × 1∘ spatial resolution) are used (ftp://ftp.whoi.edu/pub/science/oaflux/data_v3/monthly/evaporation/, last access: 1 February 2017). The TRMM (Tropical Rainfall Measuring Mission; https://pmm.nasa.gov/data-access/downloads/trmm, last access: 1 February 2017) satellite provided the precipitation information for every 0.25∘ × 0.25∘ grid on a 3-hourly to monthly timescale from 1997 to 2016 (TRMM monthly 3B43_V7 product is used). Monthly climatology of heat flux, evaporation, precipitation, and wind stress are calculated. The period of precipitation data used for climatology calculation is shorter than other parameters. The present analysis is focused on the seasonal timescale, and therefore a shorter data period will not significantly affect the results.

The daily sea level anomaly (SLA) maps are provided by AVISO (www.aviso.oceanobs.com, last access: 1 February 2017). These data are the merged product of satellite estimates from TOPEX/Poseidon, Jason-1, ERS-1/2, and Envisat and are globally available with spatial resolution of 0.25∘ × 0.25∘ from the year 1992 to present (Ducet et al., 2000; LaTraon and Dibarboure, 1999). The SLA maps are used to describe the eddy distribution in the Red Sea. The merged data from all satellite estimates provide a general picture of SLA variability and the eddy distribution in the Red Sea, even though the number of satellite tracks passing through narrow regions like the Red Sea is relatively lower than the major ocean basins. Climate Forecast System Reanalysis (CFSR; https://rda.ucar.edu/datasets/ds093.1/#!access, last access: 1 February 2017) provided an hourly wind product from 1979 to 2010 at a grid resolution of 0.312∘ × 0.312∘ (Saha et al., 2010), which is validated in the Red Sea (Aboobacker et al., 2016; Shanas et al., 2017). CFSR hourly wind at 10 m above the surface is used to study the Tokar Gap winds.

The MLD can be estimated based on different methods. Figure 2 shows a sample temperature profile collected on 19 January 2015 from the Red Sea (24.9∘ N, 35.18∘ E), with short-range gradients within the mixed layer. This gradient could rise from instrumental errors or turbulence in the upper layer. The curvature method (Lorbacher et al., 2006) identified MLD at 32 m due to the presence of a short-range gradient at this depth. The threshold method (de Boyer Montegut et al., 2004) detected the MLD at 130 m (threshold = 0.2 ∘C), while the segment method (Abdulla et al., 2016) identified the MLD at 120 m. The segment-method-based MLD could be considered as a reliable estimate compared to both the curvature (underestimation) and threshold method (overestimation). The segment method first identifies the portion of the profile with significant inhomogeneity where the transition from a homogeneous layer to an inhomogeneous layer occurs. Then, this portion of the profile is analyzed to determine the MLD (a detailed procedure of the estimation technique is given Abdulla et al., 2016). In the present study, MLD is estimated based on the segment method, which is found to be less sensitive to short-range disturbances within the mixed layer (Abdulla et al., 2016).

The availability of profiles is denser along the middle of the Red Sea during all months. The present analysis is performed for the profiles that fall within 0.5∘ to the east and west of the main axis that, running along almost the middle of the Red Sea (hereafter called the “main axis”), has the advantage of a sufficient number of profiles for every month. The main axis of the Red Sea is inclined to the west, with respect to true north, by ∼ 30∘. For this reason, instead of zonally averaging, the climatology is calculated by averaging the MLDs in an inclined direction parallel to the “cross axis” (Fig. 1). The MLD is estimated for the individual profiles, and then the monthly climatology is calculated every 0.5∘ from south to north (13 to 27.5∘ N).

The heat flux, evaporation, precipitation, and wind stress are interpolated to a 0.5∘ × 0.5∘ spatial grid to match the MLD climatology with the help of the climate data operator (CDO) tool available at http://www.mpimet.mpg.de/cdo (last access: 1 February 2017). The change in surface water buoyancy forces is calculated following Turner (1973): B0=Cp-1g∝ρ0-1Qnet+-1⋅gβsE-P=B0T+B0H, where Cp is water heat capacity, g is the acceleration due to gravity, ∝ is the thermal expansion coefficient, ρo is the density of surface water, Qnet is the net heat flux at the sea surface, β is the haline contraction coefficient, s is the salinity of surface water, E is the evaporation rate, and P is precipitation. In Eq. (1), B0T and B0H respectively represent the thermal and haline components of the buoyancy force. For ease of explanation, the Red Sea is divided into southern (13–18∘ N), central (18–23∘ N), and northern (23–28∘ N) regions and the seasons are defined as winter (December–February), spring (March–April), summer (May–August), and fall (September–November).

The Red Sea exhibits strong seasonal changes in its MLD, with deeper mixed layers during the winter and shallower ones during the summer and gradual changes from deeper to shallower and vice versa in the transitional months. A Hovmöller diagram of the monthly MLD climatology is presented in Fig. 3. The deepest MLD is observed in February and the shallowest during May–June. Significant annual variability is observed in the Red Sea. The maximum value of climatological mean MLD is observed in February in the northern Red Sea, while the minimum occurs at various instances, especially during summer months. The MLD of individual profiles in the northern Red Sea has a wide range of values from 40 to 120 m, mainly due to the presence of an active convection process, while some of the profiles show an MLD deeper than 150 m, consistent with Yao et al. (2014b).

In addition, the southern central Red Sea (14–21∘ N) also experienced deeper MLDs during winter. The observed shallow MLD patches are not considered because the noise in the MLD (∼ 44±14 m) overlaps the mean MLD of the northern (∼ 53 m) and southern (∼ 48 m) grids. The observed noise around 25∘ N is relatively small (∼30±9 m) compared to the difference in MLD values towards the northern (∼ 70 m) and southern (∼ 50 m) latitudes, and hence this is considered as a shallow MLD region.

During July to September, the region around 19∘ N experienced a deeper mixed layer in contrast with the general pattern of summer shoaling over the entire Red Sea. The deepening of the MLD begins in October throughout the Red Sea. The winter cooling and associated convection strengthens by December, with an average MLD > 50 m, which intensifies by January and persists throughout February.

The mixed layer starts to shoal gradually by the end of February, and the MLDs of most areas decrease to 20±7 m by April. Summer shoaling is comparatively stronger in the 15–18∘ N latitude band, and the detected mean MLD is < 15 m. Individual observations revealed that many profiles have MLDs < 5 m. In general, the shallow mixed layers are predominant from April to September, while this prevails until October in the far north. In the south-central Red Sea, the shallow mixed layer exists for only a short period from April to June.

MLD is directly influenced by changes in the net heat flux (NHF), freshwater flux (E-P), and wind stress. The different terms that contribute to NHF are given in Fig. 4 for a sample year 2016 in the central Red Sea. On an annual average basis, the incoming shortwave radiation (SWR; 202 W m-2, positive downward) is mainly balanced by LHF (latent heat flux, -126 W m-2) and LWR (long wave radiation, -83 W m-2), while the SHF (sensible heat flux) is only -4 W m-2. The net heat loss in the central Red Sea is 11 W m-2. Both the LHF and LWR are gradually increasing towards the northern Red Sea. The monthly climatologies of the NHF in the northern, central, and southern Red Sea are given in Fig. 5a. Heat loss rises above 200 W m-2 during December–January in the northern Red Sea, with a maximum of ∼ 250 W m-2 at the northern end of the sea in December. The annual mean of NHF is negative (heat loss) across the Red Sea, except for isolated locations in the southern Red Sea with trivial heat gain (figure not shown). The thermal components of the buoyancy forces calculated based on Eq. (1) show that the heat flux supports mixing through buoyancy loss in the northern and central Red Sea during the winter, while it opposes vertical mixing due to buoyancy gain during summer. In the southern Red Sea, the effect of heat flux is relatively weak.

The evaporation rate in the Red Sea gradually increases from south to north (Fig. 5b). The central and northern Red Sea have higher evaporation during the winter (∼ 6 mm day-1) and moderate evaporation (∼ 3 mm day-1) during the summer. Evaporation shows weak seasonality in the southern Red Sea. Precipitation in the southern region is higher than that of the other areas of the Red Sea, with maximum rainfall during July–September (Fig. 5b). The changes in buoyancy forces corresponding to freshwater flux (haline component) are estimated based on Eq. (1), which shows that the changes support vertical mixing throughout the year and over the entire Red Sea. The thermal component is relatively higher than the haline component, and the net buoyancy flux follows a more or less similar pattern of thermal buoyancy flux all along the Red Sea (figure not shown). The observed variability of the above-discussed parameters is consistent with findings from earlier studies (Albarakati and Ahmad, 2013; Sofianos et al., 2002; Tragou et al., 1999).

The pattern of wind stress in the Red Sea is significantly different from the other parameters. The wind stress is strong during the winter, leading to enhanced turbulence and mixing, while it is weak during the summer, resulting in a shallower mixed layer (Fig. 5c, d). Apart from that, strong surface winds blow to the Red Sea through the Tokar Gap at approximately 19∘ N in July and August.

The correlations between MLDs and forcing factors are given in Fig. 6. The statistical significance of the correlation values are verified based on a t test following Bretherton et al. (1999), and the estimated p value, t value, and the effective degree of freedom show that the correlation values are statistically significant at 95 %. The wind stress and E-P are positively correlated with MLD, while the NHF is negatively correlated because as NHF (into the ocean) increases, MLD decreases. For simplicity of the figure (Fig. 6), the correlation values of all parameters are presented as positive. NHF and E-P are well correlated (> 0.8) with MLD in the central and northern Red Sea and weakly correlated in the south. Wind stress has a higher correlation (> 0.8) to the south, while it is relatively weakly correlated in the central and northern Red Sea. Toward the northern end, the wind stress gradually achieves a higher correlation.

The results from Figs. 5 and 6 indicate that the MLD variability of the Red Sea is dominated by wind stress in the southern part; NHF (heat flux) and evaporation play a major role in the central region, while all three influence the northern region. Remarkably, for all the above-discussed parameters, coinciding drops are observed in the correlations at approximately 13.5, 17.5, 19, 23, and 26.5∘ N, which indicates the impact of additional forces like eddies and currents in regulating the MLD variability of the region.

Earlier studies have proved that the upper ocean is efficiently re-stratified by ocean eddies, which may significantly change the MLD. The resultant effect of eddy is largely dependent on the eddy amplitude. The mixing intensity is largest at the center of the eddy and decays on average with increasing radial distance (Dewar, 1986; Fox-Kemper et al., 2008; Hausmann et al., 2017; Smith and Marshall, 2009). The observed results show that the mixing associated with eddies dominates over the existing effect of wind stress and heat flux. CE diminishes mixing through the upwelling of the subsurface water, while AE enhances mixing through the downwelling of the surface water (de Boyer Montegut et al., 2004; Chelton et al., 2004, 2011; Dewar, 1986; Hausmann et al., 2017).

Satellite altimetry maps revealed the presence of multiple eddies in the Red Sea which are often confined to specific latitude bands (Clifford et al., 1997; Johns et al., 1999; Quadfasel and Baudner, 1993; Sofianos and Johns, 2007). Analyzing the SLA maps from 1992 to 2012, Zhan et al. (2014) reported the presence of multiple eddies with both polarities in the Red Sea. The number of identified eddies peaked at approximately 19.5 and 23.5∘ N. The upwelling proxy constructed using MODIS SST in the northern Red Sea shows the presence of frequent upwelling events at approximately 26.5∘ N almost every year (Papadopoulos et al., 2015), indicating the presence of cyclonic eddy. The extent and time of the upwelling vary from year to year. In summary, a significantly large number of eddies are noticed around 19.5, 23.5, and 26.5∘ N, which could be the possible reason for coinciding drops in the correlation around 19, 23, and 26.5∘ N.

The Red Sea is very narrow at 13.5∘ N. Moreover, complex dynamics occur in this region associated with surface and subsurface currents in the strait between the Red Sea and the Gulf of Aden. The complexity of this region prevents linking the MLD variability directly to atmospheric forcing or eddies. The region at approximately 17.5∘ N is between the two eddy-driven downwelling zones at approximately 15 and 19∘ N (Fig. 3). Mass conservation requires upwelling to replace the downwelling water. The MLD climatology shows shallow mixed layers throughout the year at 17.5∘ N, which could be due to possible upwelling. Further investigation is required to unveil the dynamics associated with this region.

The Tokar Gap is one of the largest gaps in the high orography located on the African coast of the Red Sea, near 19∘ N. Strong winds are funneled to the Red Sea through this gap which last for a few days to weeks. Figure 7a shows the u component of CFSR hourly surface wind at the Tokar region from 1996 to 2006. The figure shows that the strong wind events occur during summer every year, while the intensity and duration of the event vary from year to year. Tokar Gap winds frequently attain a speed of 15 m s-1. Previous research also shows similar results (Jiang et al., 2009; Ralston et al., 2013; Zhai and Bower, 2013). Zhai and Bower (2013) reported that wind speed may reach 20 to 25 m s-1 according to ship-based observations. Figure 7b shows that the onset of the 2001 Tokar event was on 20 July and continued until 20 August, when the maximum wind speed occurred during this period compared to rest of the year. These strong winds generate strong turbulence in the surface water, which enhances vertical mixing.

The temperature and salinity profiles measured during summer 2001 (13–14 August 2001), which coincided with the Tokar event, are shown in Fig. 8a and b (Sofianos and Johns, 2007; Zhai and Bower, 2013). The signature of the Tokar event is clearly visible in the satellite-derived SLA, with well-defined cyclonic and anticyclonic eddies to the north and south of the Tokar Gap, respectively (Fig. 8c–e). Both eddies have a basin-wide influence and radii between 70 and 80 km. The corresponding wind speed pattern (averaged for the previous 7 days) is shown (Fig. 8f–h). The profiles to the north and south of the jet axis display a significant difference in MLD, with a deeper mixed layer in the south. Station A is far from both cyclonic and anticyclonic eddies and shows the expected MLD during this period. The presence of the anticyclonic eddy at station B enhances strong downwelling, extending the mixing to a depth of approximately 80 m. It should be noted that the entire Red Sea basin is well stratified during this period, with MLDs ranging from 10 to 15 m. Stations C and D are located at the edge of the cyclonic eddy, and both have a shallower thermocline and mixed layer.

The MLDs of all the available profiles in the Tokar region before, during, just after, and after a month of the Tokar event are plotted in Fig. 9 (profiles for the first 15 days of each month are displayed). The mean MLD, standard deviation, and number of profiles are given in Table 1. Before the Tokar event, the southern and northern sides of the Tokar axis (18–19.5 and 19.5–21∘ N, respectively) displayed similar mixed layers (Fig. 9a–c). During the Tokar event, the southern side experienced enhanced mixing, while the northern side shows a shallow mixed layer (Fig. 9d–f).

The anticyclonic part of the Tokar-induced eddies enhances downwelling and the associated deepening of the mixed layer along the southern side of the jet axis, while the cyclonic eddies generate upwelling and the associated shoaling of the mixed layer along the northern side. The profiles in September (just after the Tokar event) show that the southern side is well mixed by the event, which leads to an average difference of 20 m in the MLDs between the two sides of the Tokar axis (Fig. 9g–i). The signature of the Tokar event in the MLDs (MLD difference between north and south of the jet axis) has disappeared by October (1 month after the Tokar event; Fig. 9j–l). The dominant effect of mountain gap winds on MLD changes has been reported in many studies globally, for instance in the Gulf of Tehuantepec in the eastern tropical Pacific (Gonzalez-Silvera et al., 2004; Stumpf, 1975) and Bora in the Mediterranean Sea (Grisogono and Belusic, 2009).

The mixing in the Tokar region during summer is the sum of the two mechanisms, the wind-induced turbulent mixing and the secondary circulation (eddies) induced by the wind. Both mechanisms act in the same direction on the southern side of the jet axis, resulting in enhanced mixing, while they act in the opposite direction on the northern side, leading to reduced mixing. Further studies are required for a proper quantification of the contribution of each mechanism. In summary, during the summer, the turbulence induced by strong wind and the impact of anticyclonic eddy enhance vertical mixing on the southern side of jet axis, while the wind-induced mixing is diminished by the presence of cyclonic eddy on the northern side of the jet axis.