1 Introduction

Water masses are important for climate variability and biogeochemical cycling as they transport mass, heat, oxygen, CO2, and nutrients (Cook et al. 2016; Kerr et al. 2018; Puerta et al. 2020). The water mass assessment is also used to understand the surface and subsurface circulation (e.g., Wüst 1963). Besides, changes in water mass properties can also produce steric sea level changes, as has been reported in marginal seas (e.g., Tsimplis and Rixen 2002). All these are interesting topics for the Caribbean Sea as it is connected to the North Atlantic Subtropical Gyre Circulation and the North Atlantic Western Boundary Current System, which is linked to the Atlantic Meridional Overturning Circulation (AMOC) (Candela et al. 2019; Cunningham et al. 2007; Tuchen et al. 2022).

According to Andrade (2000), water masses in the Caribbean Sea include the Caribbean Surface Water (CSW), characterized by its low salinity due to river outflow and precipitation in the region. The North Atlantic Subtropical Underwater (SUW) forms in the North Atlantic Subtropical Gyre and is characterized by its high salinity. The Antarctic Intermediate Water (AAIW) and North Atlantic Deep Water (NADW) enter the Caribbean Sea through the deeper sills in the Antilles (Table Aux. 1). The former enters through the Granada Basin and spreads northwestward, entering the Cayman Sea, where it is separable by its salinity minimum, whereas the NADW fills the deep Caribbean Sea. This agrees with other water mass studies in the Caribbean Sea (Casanova-Masjoan et al. 2018; Gordon 1967; Hernández-Guerra and Joyce 2000; Morrison and Nowlin Jr., 1982; Smith 2010; Wüst 1963). However, other water mass classifications can be found in literature (e.g., South Atlantic Water) (Rhein et al. 2005), whose characteristics are comparable to the water masses previously mentioned.

The Caribbean Current transports surface water masses from the North Atlantic Ocean, entering the Caribbean Sea through the Antilles arc. Seasonal air–sea interaction processes modify them locally (Beier et al. 2017; Montoya-Sánchez et al. 2018a). Besides, Correa-Ramirez et al. (2020) suggested that the SUW maximum salinity core (placed at ~ 100–150 m depth) was diluted by mechanical mixing with fresher surface waters and by double diffusion with the western North Atlantic Central Water (wNACW) below (~ 300 m deep). wNACW results from the mixing between SUW and AAIW; therefore, we do not include its assessment in this work.

Furthermore, the southward flow of the AMOC lower limb was estimated by the sum of the upper and lower NADW transport. The former (UNADW) occupies the 1100–3000 m layer (Cunningham et al. 2007) and would be the deeper water mass entering the Caribbean Sea due to topographic constraints. Labrador Sea Water has also been reported to enter the Caribbean Sea through the deeper sills (Joyce et al. 1999); however, this water mass has been associated to the UNADW (MacCready et al. 1999).

Almeida et al. (2018) assessed water masses related to the AMOC in the Atlantic Ocean and updated their particular hydrographic characteristics in terms of the Thermodynamic Equation of Seawater 2010 (hereinafter TEOS-10), including the AAIW (2.04–5.98 °C/33.97–34.98 g kg–1) and NADW (1.49–3.98 °C/34.95–35.16 g kg–1). The reported water mass conservative temperature (Θ) and absolute salinity (SA) ranges are large, as they are used to describe water mass properties present in the entire Atlantic basin.

In this work, we revisit water masses in the Caribbean Sea, studying in depth several aspects that have not been addressed before. We first assess the CSW seasonal variability from a climatological approach, studying its properties at the mixed layer depth (MLD). Large seasonal variations result from a combination between changes in advected waters entering the Caribbean Sea and local air–sea processes. Furthermore, we offer a new perspective related to the interaction between the fresher CSW and saltier SUW, which produces a vertical density stratification controlled by salinity stratification (barrier layer) that varies seasonally, modulating the static stability with important consequences for local ocean dynamics.

Second, we assess the individual volume and spatial distribution of subsurface, intermediate, and deep-water masses in the Caribbean Sea, using hydrographic data from a climatological annual mean. This is done in the study area based on water mass fractions. To achieve this goal, we define the water masses particular characteristics (conservative temperature, absolute salinity, and dissolved oxygen) when entering the Caribbean Sea, what has not been previously done using TEOS-10. These results are important as they serve as a climatological starting point for future studies to assess changes in volume and properties of these water masses at different time scales (e.g., Andrade 2000; Masson-Delmotte et al. 2021). Such temporal changes in water masses will not only modulate the transport of water properties such as heat or CO2, but can also produce a signature in steric sea level, which in the long-term will be associated to climate warming trends (Silvy et al. 2020). Understanding future mean sea level rise in the Caribbean Sea is of main importance, due to the large expected impacts in the low-lying coastal areas (Nicholls and Cazenave 2010).

Third, we assess sub-annual variations in intermediate and deep-water masses with respect to the climatological approach previously shown. For this, we used eight temperature and salinity time series collected from April 2007 to February 2008 at different depths from three moorings placed in the Guajira upwelling system (Fig. 1). These novel results show large sub-annual variations in the Caribbean Sea water masses at 500 m, which are reduced with depth.

Fig. 1
figure 1

Map of the Caribbean Sea indicating basins and most important geographical features and passages to the Atlantic Ocean. Gray contours show 200-, 1800-, and 3000-m isobaths. The World Ocean Atlas nodes used to assess climatological hydrography are shown in orange. Mooring locations where observations were collected are indicated in purple. Main rivers are shown in blue

Full size image

As a final step, we take advantage of these time series and assess the Guajira upwelling dynamics. We found a relation between the rotated wind stress and temporal variations in water mass properties, regularly down to 700 m depth and occasionally deeper, indicating deep-water advection shoreward across the continental margin. To our knowledge, such deep effect of the Guajira upwelling has not been previously reported. Furthermore, our results show that temporal variations in intermediate and deep-water mass properties can be used to assess upwelling dynamics, what is useful as direct observation of vertical currents are difficult to measure (Pond and Pickard 1983).

2 Study area

The Caribbean Sea is placed in the intertropical zone between 7.8°N and 23°N, with zonal limits between the 60°W and 90°W (Fig. 1). The flow of Atlantic water into the Caribbean Sea through the different passages has been studied to understand their connection (Fratantoni et al. 2000; Wilson and Johns 1997). According to Johns et al. (2002), nine passages have sill depths > 400 m in the Antilles arc between Cuba and South America, which allow Atlantic inflow into the Caribbean Sea (Table Aux. 1). From a total estimated inflow from the Atlantic into the Caribbean Sea of ~ 28 Sv (1 Sv = 1 × 106 m3s–1), about 17 Sv are wind-driven from the North Atlantic Subtropical Gyre Circulation, mostly entering north of ~ 15° N. The additional 11 Sv seems to enter through the Grenada, St. Vincent, and St. Lucia passages (Windward Islands), forced by the AMOC (Johns et al. 2002). Therefore, surface and subsurface waters are expected to enter the Caribbean Sea forced by the wind-driven subtropical gyre circulation, while intermediate and deep water would inflow into the Caribbean Sea driven by thermohaline circulation (Candela et al. 2019; Tuchen et al. 2022; Vallès-Casanova et al. 2022).

The atmospheric seasonal variability in the Caribbean is controlled by the intertropical convergence zone (ITCZ) position, which controls wind and precipitation (Poveda et al. 2006). From December to April, the ITCZ meridional position is toward the south of the basin, allowing the northern trade winds to intensify and reducing precipitation. From August to November, the warmer months of the year coincide with the ITCZ meridional position toward the north of the Caribbean Sea. This enhances precipitation and weakens the northern trade winds, allowing the southern trade winds to reach the south of the Colombia Basin. These are known as the dry–windy season and the rainy–warm season, respectively (Andrade 2000). In July to August, a relative reduction in precipitation and intensification of wind in the Caribbean is known as the mid-summer drought, caused by the intensification and expansion of the North Atlantic subtropical high, forcing more intense northern trade winds in the Caribbean (Gamble and Curtis 2008). The Caribbean low-level jet dominates the north trade winds in the center of the Eastern Caribbean (~ 15° N, 75° W) (Amador 2008). This westerly near-surface jet placed at 925 hPa has two maximum speeds in January and July, thus being affected by the ITCZ position and mid-summer drought (Wang and Lee 2007).

The Caribbean Current dominates the Caribbean Sea with influence down to ~ 700 m depth (Andrade 2000). It flows to the west from the Lesser Antilles, entering the Cayman Sea and leaving the basin through the Yucatan Channel (Jouanno et al. 2008; Richardson 2005). The Caribbean low-level jet impulse and direction, parallel to the northern South America coasts, produce the Southern Caribbean coastal upwelling system (which includes the Guajira Upwelling System) (Andrade and Barton 2005; Gordon 1967). The upwelling is stronger in February (windy season) and weaker near October (rainy season). The upwelling signal includes the sea surface temperature reduction, salinity increase, and lower sea level toward the coast (e.g., Correa-Ramirez et al. 2020; Montoya-Sánchez et al. 2018b; Torres and Tsimplis 2012). The cyclonic Panama–Colombia Gyre dominates the circulation in the south of the Colombia Basin (Andrade and Barton 2000; Richardson 2005; Ruiz-Ochoa et al. 2012). The wind stress curl forces this gyre (Torres and Tsimplis 2012). Andrade et al. (2003) reported the Panama–Colombia Countercurrent as a subsurface northeastern flow along the Colombia continental coast, detaching from the eastern side of the Panama–Colombia Gyre. The countercurrent deepens to ~ 100 m when reaching the Guajira coast and continues eastward along the Venezuela coast until it leaves the Caribbean Sea through the Granada Passage.

Mesoscale cyclonic and anticyclonic eddies modulate the circulation in the Caribbean Sea. Baroclinic eddies in the region vary in diameter (200 to 500 km), frequency of occurrence (50 to 110 days), and propagation speed (12 to 15 cm s−1), while they intensify when moving westward from the Lesser Antilles into the Colombia basin (Jouanno et al. 2008; van der Boog et al. 2019; van Westen et al. 2018). Eddies are important in the Caribbean Sea for the advection and mixing of different source waters, mixing coastal with offshore waters, and forcing the circulation at deeper layers (Richardson 2005).

The freshwater plume in the Granada and Venezuela Basins originates from both the Orinoco (mean outflow of ~ 3.9 × 104 m3s–1) and the Amazon rivers, with minimum salinity in September to October and maximum in March (Chérubin and Richardson 2007; Muller-Karger and Castro 1994). The Magdalena River (Fig. 1) is responsible for the largest freshwater supply into the Caribbean Sea, with a mean flow of ~ 6.5 × 103 m3s–1 (Restrepo et al. 2014). According to Beier et al. (2017), the total annual mean river discharge into the Colombia Basin, which drains mostly south of 12° N, is ~ 50% of the annual mean discharge of the Orinoco River into the Eastern Caribbean, with a minimum flow in March and maximum in November. Besides, they show that in this basin, precipitation and evaporation terms are about five times higher than the river inflow term. Furthermore, the salt balance is in equilibrium with a strong seasonal variability, where in the dry season, it is a concentration basin (net freshwater flux out from the ocean), whereas it is a dilution basin (net freshwater flux into the ocean) for the rest of the year (Beier et al. 2017; Yoo and Carton 1990).

3 Data and methods

The World Ocean Atlas 2018 (WOA18) was used to obtain temperature (Locarnini et al. 2018), salinity (Zweng et al. 2018), and dissolved oxygen (Garcia et al. 2018) at standard depths in the Caribbean Sea (https://www.nodc.noaa.gov/OC5/woa18/). WOA18 generates climatological information by objective analysis of a wide range of scientifically quality-controlled historical data, obtained from observations gathered from different sources.

We use TEOS-10 to derive conservative temperature (Θ), absolute salinity (SA), and potential density anomaly (σ0) using a thermodynamic approach. This improved the accuracy of calculating water density allowing a more realistic analysis of physical processes such as circulation and mixing processes (IOC et al. 2010). We used the GSW toolbox version 3.06 (McDougall and Barker 2011). Surface and hydrographic sections figures use color schemes from the Cmocean toolbox (Thyng et al. 2016).

To describe the seafloor topography and determine most important passages sills’ depths, we use the General Bathymetric Chart of the Oceans (GEBCO) with a resolution of 15 s of arc, thus a spatial resolution < 500 m (GEBCO Bathymetric Compilation Group 2019).

3.1 Water masses in the mixed layer depth (MLD) and at the salinity maximum

To assess surface water mass variations, we used conservative temperature and absolute salinity values at the mixed layer depth (MLD), derived from the WOA18 quarter-degree grid. We used the annual mean, as well as the mean for March and November, to observe contrasting conditions in the hydrography from the dry and rainy seasons, respectively. Besides, these months were selected as they are at the end of these seasons, which coincides with the river runoff minimum and maximum (e.g., Beier et al. 2017), what affects the salinity at the MLD.

To estimate the MLD base, we use a potential density threshold value (0.03 kg m–3) from the 10 m depth value (de Boyer Montégut et al. 2004) which has been previously used in the Caribbean Sea (Jouanno and Sheinbaum, 2013). This reference value avoids a large part of the strong diurnal cycle (especially in equatorial regions due to wind and solar warming), affecting the top few meters of the ocean. Besides, it avoids measurements close to the surface where instrumental noise is larger. We also calculated the base of the isothermal layer depth. It was calculated from a conservative temperature difference of 0.2 °C from the 10 m depth value. A shallow bias of 25% is expected as we calculate the isothermal layer from averaged profiles (de Boyer Montégut et al. 2004). In this way, we assure to obtain the heat and salinity content of surface waters, regardless of reporting a shallower base of the MLD or isothermal layer.

Consequently, we define surface water masses as those above the pycnocline, where water properties are vertically uniform due to air–sea fluxes, mechanical inertia, lateral advection, and ocean stratification. Therefore, MLD will change in diurnal, seasonal, and intra-annual time scales. The same dataset was used to assess variations from the highest absolute salinity value in the water column below 70 m depth, which distinguishes the SUW core. This depth threshold was established to avoid shallow areas where salinity does not represent a dominant SUW presence. We use the core method to assess surface and subsurface circulation (Wüst 1963).

Static stability measures changes of density with depth. We compute the static stability (\(E\)) from the density (\(\rho\)) and depth (\(z\)) differences between the MLD and the highest absolute salinity layer (SUW core) using the equation \(E=-(1/\rho )(d\rho /dz)\) (Stewart 2008).

3.2 Water mass spatial distribution below the salinity maximum

The optimum multiparameter (OMP) inverse method (Tomczak and Large 1989) has been successfully applied at basin and sub-basin scales to infer water mass circulation and mixing (Almeida et al. 2018; Torres Parra et al. 2020). The basic assumption of OMP analysis is that all water masses can be represented as mixtures of a few water types. A water type is a point in parameter space (each hydrographic property has a single value, as the end points of its TS diagram). Besides, a source water type is a point in parameter space that belongs to the functional relationship of a water mass. Therefore, the OMP is based on a linear mixing model, so that the water fraction distribution depends on the source water types selected. The number of water types resolved by the OMP analysis is restricted to the number of hydrographic properties available (Poole and Tomczak 1999).

We performed the OMP analysis to quantify SUW, AAIW, and NADW contributions to the oceanic volume in the Caribbean Sea and their mixture fractions below the salinity maximum, allowing us to determine their spreading and mixing in the basin. Therefore, we used WOA18 hydrographic data below the salinity maximum in the Caribbean Sea with at least 70 m depth to avoid shallower nodes. Results are presented as 3D surface plots for the Caribbean Sea and in two depth distance hydrographic sections along selected transects.

The first step in the OMP analysis process is to represent the water masses by source water types. We used conservative temperature, absolute salinity, and dissolved oxygen derived from WOA18 one-degree grid to define the three source water type’s indices. Dissolved oxygen is only available at this spatial resolution due to less available data for the objective analysis (Garcia et al. 2018). Monthly mean data, available to 1500 m depth, was used to determine the SUW and AAIW source water-type indices, as at these depths, seasonal changes can be important. Annual mean data, available to 5500 m depth, was used to determine the NADW source water index, as it enters the Caribbean Sea through the deeper sills in the Antilles (Table Aux. 1), below 1500 m depth.

We estimated our own OMP parameter weights following Tomczak and Large (1989), using the variance of the source water-type indices (Table 1) and the variance of the WOA18 samples used in the OMP analysis in the Caribbean Sea. We assigned the largest parameter weight obtained to mass conservation. Mass conservation residual mean was < 1%, less than the normally accepted threshold (< 10%), indicating that the water samples were well represented by the selected source water types (e.g., Poole and Tomczak 1999). We conducted a sensitivity assessment of the OMP results based on changing parameter weights computed from data subsets. OMP mixture fractions did not show significant variations, and mass conservation residuals did not increase, giving additional confidence in the results presented herein.

Table 1 Source water-type indices and weights used in the optimum multiparameter analysisa
Full size table

3.3 Water mass sub-annual variability and relation to the Guajira upwelling dynamics

A program to characterize the oceanographic conditions in the Colombian northern continental margin was executed by Petrobras, Woods Hole Group, and Colombia’s Maritime Authority — DIMAR — as part of an offshore oil exploration program (Moreira and Sartori 2016). The Colombian R/V “ARC Malpelo” applied mooring installation and maintenance in five cruisers at three locations (Fig. 1): F1 (11.313° N, 74.566° W), F2 (11.707° N, 73.672° W), and F3 (12.812° N, 71.675° W). Conductivity, temperature, and pressure time series were collected every 5 min from April 2007 to February 2008 at 500 and 700 m depth at all the moorings. Besides, at F2, data were also collected at a depth of 1200 and 1450 m. The deeper instruments were placed 50 m above the bottom. The time series are divided into four segments, as moorings were recovered for maintenance and placed back in the same place after few hours. A SBE–37–SMP was used with an accuracy of ± 0.002 °C in temperature, ± 0.0003 mS cm–1 for conductivity and ± 0.1% of the pressure range. Instruments were calibrated before and after the project showing good performance. Data were carefully inspected for quality control, and outliers were removed. When moorings were deployed, pressure data showed an oscillation of the instruments, which was removed, keeping the data once the instruments were stable at the depth of deployment. A single daily time series was constructed using the four segments after quality control.

Zonal and meridional wind at 10 m height were obtained from ERA5 with a 6-h temporal resolution and a 0.25° spatial resolution for the 2007–2008 period (Hersbach et al. 2020). Time series were obtained from the nearest node to each mooring. Daily time series were computed averaging all the available values for each day.

We also performed the OMP linear mixing model to the mooring time series (e.g., Leffanue and Tomczak 2004). As dissolved oxygen was not available, only two parameters were used (Θ and SA), allowing us to assess the mixing of two water types at the same time (Section 3.2). Daily time series of water mass fractions were obtained from this analysis.

We use a linear regression fitted to the daily mooring Θ and SA time series to compute the residual power spectrum and report the more energetic frequencies after we removed the mean and trend. We applied the same procedure to the time series of water mass fractions obtained from the OMP analysis to the in situ data.

Besides, the low-frequency relation between water mass fractions and the rotated surface wind stress was assessed, as the latter forces local upwelling in the region (e.g., Torres and Tsimplis 2012). We rotated the zonal and meridional wind following the procedure used by Orfila et al. (2021). However, we did not use a fixed angle for the time series at the three moorings. In each case, we make to coincide the zonal wind component (U) with the most frequent wind direction. We used this procedure because coastal upwelling is due to Ekman transport, which represents displaced water from an averaged layer in a direction to the right of the wind in the northern hemisphere. The inspection of the wind rises in the 2007–2008 period (divided in 36 directions) showed a steady direction mainly between the northeast and the east in all moorings. Due to the coastline orientation in the Guajira Peninsula (Fig. 1), a wind blowing from any of these directions would produce an offshore surface water displacement, forcing coastal upwelling.

All zonal wind components were rotated in an angle measured counterclockwise from the east, so 0° indicates wind blowing from the east to the west; besides, the relative data intensity in each direction is indicated. In F1, U was rotated 30° (33.7% followed by 32.2% at 40°). In F2, U was rotated 20° (42.7% followed by 30.2% at 10°). U was rotated 10° in F3 (56.6% followed by 18.7% at 0°). The rotated zonal wind (\({U}_{R}\)) mean is negative in all moorings, indicating that the wind blows mainly toward the west to southwest (northern trade winds’ main direction). Wind stress (\(\tau =-{\rho }_{a}{C}_{D}{U}_{R}\left|{U}_{R}\right|\)) was calculated from the rotated zonal wind, drag coefficient (\({C}_{D}=1.4x{10}^{-3}\)), and air density (\({\rho }_{a}=1.3 kg {m}^{3}\)) to assess the horizontal force of the wind on the sea surface. We reverse \(\tau\) sign, to make coincide wind stress intensification with stronger upwelling, as negative values of \({U}_{R}\) enhance Ekman transport offshore.

A 20-day low-pass filter was applied to the daily time series of rotated wind stress and water mass fractions obtained from the mooring observations. Besides, low-frequency time series were standardized by removing their mean and dividing them by the standard deviation. We assessed a causal relationship between these daily time series evaluating their correlation coefficient with a significance level of p < 0.01. Besides, lagged correlations were explored moving the series up to ± 30 days.

4 Results and discussion

4.1 CSW seasonal variability and interaction with SUW

Seasonal variations of the CSW are studied with the hydrographic conditions at the MLD. The annual MLD based on a difference of 0.03 kg m−3 with respect to the potential density at 10 m depth is inexistent at the south of the Eastern Caribbean, thus reported as being 10 m depth (Fig. 2a). Toward the north of the Eastern Caribbean and in the Cayman Sea, the MLD is between 20 and 40 m depth. The lack of a deeper mixed layer is related to the presence of a barrier layer in the Caribbean Sea throughout the year. A barrier layer occurs when the MLD (based on density homogeneity) is shallower than the isothermal layer (de Boyer Montégut et al. 2004; Sprintall and Tomczak 1992). The isothermal layer depth in the Caribbean Sea is about twice the density derived MLD (Fig. Aux. 1). In the Caribbean Sea, the barrier layer is caused by a strong halocline connecting diluted surface waters (CSW) with saltier subsurface waters (SUW), which increases density along a deeper isothermal layer (Fig. Aux. 2). Therefore, vertical density stratification (static stability) in the Caribbean upper ocean is controlled by salinity.

Fig. 2
figure 2

Mixed layer depth (MLD) based on density criterion from WOA18 annual (a), March (b), and November (c) means. Isobaths 10, 20, and 50 m. Annual (d), March (e), and November (f) conservative temperature at the MLD. Isotherms 26, 27, 27.5, 27.8, and 28.5 °C. Annual (g), March (h), and November (i) absolute salinity at the MLD. Isohalines 35.7, 36, and 36.3 g kg−1. Annual (j), March (k), and November (l) maximum absolute salinity below 70 m depth indicates SUW core. Isohalines (red) 37 and 37.2 g kg −1. Isobaths (black) 100, 150, and 200 m indicate the maximum salinity depth. Annual (m), March (n), and November (o) static stability between the MLD and SUW core. Isolines of 1.5, 2, and 2.5 × 10−5 m−1

Full size image

The annual temperature at the MLD ranges between 25 and 29 °C (Fig. 2d), while the annual salinity ranges between 35 and 36.5 g kg−1 (Fig. 2g). Lower temperatures and higher salinity denote the Southern Caribbean Upwelling System. High temperature and low salinity are found in two places: first, in the Darien and Mosquito gulfs, as it is a dilution sub-basin (Beier et al. 2017), and second, diluted waters in the Venezuela and Granada basin mixed layer (< 35.7 g kg−1) denote the Orinoco freshwater plume entering through the Lesser Antilles. As these waters are transported westward by the Caribbean Current, they increase their salinity due to horizontal and vertical mixing and by local air–sea fluxes. In both cases, diluted waters coincide with warmer waters, probably due to a strong pycnocline which enhances vertical stratification, limiting the absorption of solar radiation to a thin surface layer (Ruiz-Ochoa et al. 2012) and reducing the inflow of colder subsurface water into the mixed layer (e.g., Ke-xin and Fei 2021).

In March, at the end of the dry–windy season, which coincides with the end of the boreal winter, temperature in the mixed layer drops about 1 °C in the entire Caribbean Sea (Fig. 2e), when compared to the annual mean. In addition, as the Orinoco’s minimum flow is in March, the salinity increases in the Eastern Caribbean as shown by the 35.7 g kg−1 isohaline, which does not enter the basin (Fig. 2h). The Mosquito-Darien dilution sub-basin is still distinguishable, regardless of the precipitation and river outflow decrease in March. Due to stronger winds, the upwelling strengths depict a larger area compared to the annual mean, where salinity increases and temperature decreases. An increase of density in the MLD (colder and saltier waters) reduces stratification (Fig. 2n), which, together with wind intensification, produces a deeper mixed layer (Fig. 2b).

In November, the end of the rainy season coincides with the Orinoco and other rivers in the Eastern Caribbean maximum freshwater outflow, decreasing the MLD salinity in the entire basin (Fig. 2i) when compared to the annual mean. Diluted waters from the Orinoco River (< 35.7 g kg−1) enter through the Lesser Antilles and reach the Central America Rise south of Jamaica. Besides, the Mosquito-Darien dilution sub-basin is noted by salinities < 35.7 g kg−1. Air–sea heat fluxes, together with a stronger surface stratification (Fig. 2o) and weaker winds, act together to increase the MLD temperature (> 27.8 °C) in most of the Caribbean Sea (Fig. 2f), weaken the southern Caribbean upwelling and decrease the MLD in the Eastern Caribbean (Fig. 2c).

Therefore, the mean CSW hydrography depicts large temperature and salinity ranges (Fig. 3) related to basin spatial variations. Cold salty waters characterize the southern Caribbean upwelling area. In contrast, warm diluted waters depict the Orinoco river plume and the Mosquito-Darien sub-basin. In addition, CSW hydrography is characterized by large seasonal changes. Warmer diluted waters dominate the rainy season from August to November, while relatively colder and saltier waters dominate the windy season from December to March.

Fig. 3
figure 3

Conservative temperature (°C) and absolute salinity (g kg−1) diagram. Annual means for seven WOA18 nodes in the Caribbean Sea are shown at standard depths using different colors. Nodes’ location are shown in Fig. 1. Black circles indicate modified source water type indices used for OMP analysis: North Atlantic Subtropical Underwater (mSUW), Antarctic Intermediate Water (mAAIW), and North Atlantic Deep Water (mNADW). The presence of the CSW and western North Atlantic Central Waters (NACW) is indicated with their abbreviation. Light gray lines show potential density anomalies (kg m−3). Circles and numbers with the node’s color indicate the depth at which the maximum salinity (above 26 kg m−3) or minimum salinity (below 27 kg m−3) occur, indicating the SUW and AAIW core, respectively. Dashed square indicates the area shown in Fig. 6

Full size image

Montoya-Sánchez et al. (2018a) assessed seasonal variability of the mixed layer heat budget in the Caribbean Sea, computing the MLD based on density variations from the 10 m depth reference value but only driven by potential temperature changes. Therefore, they did not explore the barrier layer caused by vertical salinity stratification as we do. Consequently, they found a relatively shallow MLD oscillating spatially between 10 and 90 m, what is similar to the isothermal layer depth we report (Fig. Aux. 1). In their study, they show the relation between MLD variability and air–sea fluxes, including the Caribbean low-level jet wind stress and surface heat flux. Besides, they show the shallower MLD in the rainy season and off the coast of Central and South America year round. Therefore, their study agrees with the hydrographic results we present at the MLD.

To assess variations in the barrier layer, we study salinity differences between the MLD and the SUW core. The larger these salinity differences, the stronger the halocline and therefore, the more stable the interface (larger static stability). The SUW core can be tracked by its highest salinity in the water column (Fig. 3). SUW enters the Caribbean Sea through the passages east of Puerto Rico and north of St Vincent Passage, as indicated by the 37.2 g kg−1 isohaline (Fig. 2j). SUW core dilutes as it continues westward transported by the Caribbean Current, reaching the Yucatan Chanel. In all the Caribbean Sea, salinity at the MLD is lower than at the subsurface, with larger differences (> 1.5 g kg−1) in the Granada Basin due to the contrast between most diluted surface waters by the Orinoco plume and the saltier subsurface waters from the SUW core entering the basin. Smaller differences (< 0.7 g kg−1) are found in the Southern Caribbean coastal Upwelling System and Cayman Sea.

The SUW core enters through the northeast Caribbean Sea boundary at ~ 150 m depth as it is denser (~ 25.5 kg m−3 in Fig. 3) than surface waters. SUW core is found at shallower depths (~ 100 m) toward the south of the Eastern Caribbean and deeper than 150 m north of ~ 15° N (Fig. 2j). This dipole pattern results from the Caribbean low-level jet wind stress curl, forcing convergence (divergence) of surface waters toward the Greater Antilles (south Colombia Basin) (Montoya-Sánchez et al. 2018a). This meridional depth gradient in the SUW core coincides with the MLD spatial variation (Fig. 2a).

SUW core does not have large seasonal variations in its salinity and depth when compared to CSW at the MLD (Fig. 2). Therefore, static stability in the Caribbean Sea, which is dominated by salinity stratification (barrier layer), is dominated by surface salinity variations. The mean annual static stability is strong in the Caribbean Sea (> 1 × 10−5 m−1 in Fig. 2m). It is the largest in the Granada Basin and the south of the Colombia Basin, where fresher conditions are found in the MLD, and it reduces to the northwest where saltier conditions in the MLD prevail. Close to the Guajira Peninsula, vertical mixing forced by upwelling reduces the static stability, when compared to nearby coastal areas. In March (dry season), when salinity at the MLD increases, static stability reduces; on the contrary, during November (rainy season), when fresher waters are present in the MLD, the upper ocean is more stable (Fig. 2n and o). Therefore, static stability in the Caribbean upper ocean has large spatial and seasonal variations.

4.2 Volume and spatial distribution of subsurface, intermediate, and deep-water masses

4.2.1 Source water type indices and Θ–SA diagram

The OMP analysis is used to assess volume and spatial distribution of the three water masses found below the salinity maximum. To perform this analysis, source water indices must be defined for the SUW, AAIW, and NADW, what we do in terms of their conservative temperature, absolute salinity, and dissolved oxygen. Besides, we need to differentiate the specific source water indices of these water masses when entering the Caribbean Sea, from their hydrographic characteristics as can be found in the Atlantic. In this ocean, these water masses are characterized by large hydrographic ranges, as their properties erode with advection though a broad region (e.g., Almeida et al. 2018). Therefore, for this paper, we use mSUW, mAAIW, and mNADW as specific “modified source water type indices,” to assess the presence of SUW, AAIW, and NADW in the Caribbean Sea. Defining particular source water indices for the Caribbean Sea is important, as their use allows for an accurate evaluation of water mass proportions and distributions in this basin (e.g., Tomczak and Large 1989).

To determine the modified source water-type indices, we assessed hydrographic conditions at the WOA18 nodes close to the Antilles passages, through which water masses enter into the Caribbean Sea. SUW is part of the North Atlantic Subtropical Gyre, and its core is represented by the highest salinity in the water column. The highest salinity from all the monthly means is found in January at 150 m depth entering the Caribbean Sea east of Puerto Rico (node placed at 17.5° N to 65.5° W). AAIW core is represented by the lowest salinity below the depth of maximum salinity. The lowest salinity from all the monthly means is found in January at 700 m depth entering the Caribbean Sea through the Windward Islands (St. Vincent and St. Lucia Passages, node placed at 13.5° N to 61.5° W). NADW flows through the deep Anegada Passage into the Caribbean Sea (e.g., MacCready et al. 1999). To define this modified source water index, we use hydrographic properties from the closest deep node to the Anegada Passage inside the Caribbean Sea (17.5° N to 64.5° W) at is sill’s depth (~ 1850 m in Table Aux. 1). These indices are shown in Table 1.

The defined source water indices correctly represent water mass mixture in the Caribbean Sea as seen in Θ–SA diagram (Fig. 3), where the annual hydrographic mean of seven WOA18 nodes is shown (Fig. 1). The CSW can be found at densities < 24.5 kg m−3. Its temperature is > 25 °C and has a salinity range of 35–36.5 g kg−1. CSW mixes with SUW, whose highest salinity core is reached between 95 and 200 m depth, with a temperature of ~ 23 °C and density of ~ 25.5 kg m−3. The linear relationship connecting 26 to 27.3 kg m−3 in the Θ–SA diagram indicates the mixture of the saltier–warmer SUW with the diluted colder AAIW. AAIW core has a density of 27.3 kg m−3 and is present between 700 and 850 m depth. From this depth, water increases its salinity and reduces its temperature until it reaches the NADW characteristics with a density of ~ 27.75 kg m−3, which fills the deeper Caribbean Sea.

Water mass distribution in the Θ–SA diagram supports the SUW core advection shown in Fig. 2j, entering through the northern Lesser Antilles passages, which agrees with previous studies (Montes et al. 2016; Shcherbina et al. 2015). Anegada node has the most similar water properties to the Θ–SA pair we use as source water index (Fig. 3). SUW spreads south and westward. Its core sinks toward the nodes in the Colombia and Cayman Basins, while it rises toward the nodes in St Lucia, Guajira, and Darien. The Guajira node shows the most diluted SUW core as a consequence of the upwelling forced vertical mixing. AAIW core with the lowest salinity is at St Lucia node as it enters into the Caribbean Sea through the Windward passages (Fig. Aux. 3), thus closest to the Θ–SA pair we use as source water index. The saltier AAIW core is at the Cayman node as it mixes in its transit to the northwest, with a difference of ~ 0.15 g kg−1.

Water property profiles at the seven WOA18 nodes assessed (Fig. Aux. 2) show in all cases that surface salinity is lower than subsurface salinity maximum (> 36.5 g kg−1). Therefore, the strong salinity stratification (barrier layer) produces a surface halocline present in all nodes. Besides, higher surface density occurs in Guajira node due to the colder and saltier upwelled water. Surface dissolved oxygen is similar in all nodes (~ 200 µmol kg−1). Below ~ 50 m, dissolved oxygen gently reduces until it reaches its minimum at ~ 600 m indicating the presence of AAIW, characterized by its low dissolved oxygen content (< 140 µmol kg−1). Below ~ 600 m, dissolved oxygen increases reaching over 200 µmol kg−1 at 1400 m, due to the presence of the more ventilated NADW (Table 1). Water properties in the Eastern Caribbean Sea using WOA18 data (Fig. 3) agree to what has been described from particular hydrographic observations in the basin (e.g., Correa-Ramirez et al. 2020).

4.2.2 Water mass distribution below the salinity maximum

The OMP analysis was performed to the mean annual hydrographic values of 190 WOA18 nodes in a layer from the salinity maximum down to the bottom, thus avoiding surface water masses. Thirty-eight of the 228 one-degree nodes inside the Caribbean Sea were removed for being shallower than 70 m. The OMP analysis indicates that 46% of the volume in this layer corresponds to NADW, 38% to AAIW, and 16% to SUW. Above the maximum salinity depth, a mixture of CSW and SUW is not included in these water fractions.

Maximum SUW core fractions are found close to the north–eastward boundary in the Caribbean Sea at ~ 100 m depth (Fig. 4a), as this limit is closest to the North Atlantic Subtropical Gyre, from which this water type comes from. As SUW is advected westward by the Caribbean Current, its core is diluted but keeping a concentration > 85%. Its core deepens toward the northwest (Fig. 5a) while surges toward the south (Fig. 5b) of the Eastern Caribbean. This spatial distribution is consistent with the location of the maximum salinity core (Fig. 2j). SUW layer thickness (water fraction ≥ 30% from below the maximum salinity) extends from 150 to 450 m (Fig. 4d). SUW has the thinner layer thickness of the three water masses shown in Fig. 4.

Fig. 4
figure 4

Water-type maximum fraction (color) and depth (isobaths in m) at which it is located from OMP analysis: (a) SUW; (b) AAIW; and (c) NADW. Vertical extent of the water-type layer (in m) with fractions > 30% for (d) SUW; (e) AAIW; and (f) NADW. In panel (d), red “x” indicates location of nodes in hydrographic section 1 and red “ + ” nodes in hydrographic section 2

Full size image
Fig. 5
figure 5

Water-type fractions from OMP analysis: SUW (a) in hydrographic section l and (b) in hydrographic section 2. AAIW (c) in hydrographic section 1 and (d) in hydrographic section 2. NADW (e) in hydrographic section l and (f) in hydrographic section 2. Only contributions > 30% are shown. Bottom along the hydrographic section shown by gray shading. Each water type has a different depth range. Figure 4d shows hydrographic sections location

Full size image

AAIW is present in all the Caribbean Sea with maximum water fractions > 90%; its core is 650 to 800 m depth, except at shallow areas (Fig. 4b). Maximum AAIW core concentrations are found toward the southeastern Caribbean Sea boundary because it is where AAIW enters the basin (e.g., Rhein et al. 2005). Lower AAIW core concentration is found toward the Cayman Sea due to mixing as it is advected northwest by the Caribbean Current. Consequently, there is a larger AAIW volume in the Eastern Caribbean than in the Cayman Sea, as indicated by a thicker AAIW layer in the former (~ 1100 m) than in the latter (~ 900 m) (Fig. 4e). These differences in the AAIW layer vertical extent can also be seen in hydrographic sections 1 and 2 (Fig. 5c and d). Note how the shallower AAIW limit deepens to the west and the north, respectively, in response to the deepening of the SUW core (Fig. 5a–b). Topography of the Central America Rise does not severely constrain AAIW advection from the Eastern Caribbean into the Cayman Sea (Fig. 1), as the Jamaica and Pedro–Rosalinda Channels are 1540 and 1280 m depth (Table Aux. 1). Gray shading in Fig. 5c denotes the Pedro–Rosalinda sill. The spatial distribution of AAIW fractions in the Caribbean Sea, based on OMP analysis, coincides with the annual spatial distribution of the minimum salinity in the water column (Fig. Aux. 3), allowing us to observe a small seasonal variability.

The OMP analysis indicates that maximum NADW fractions (~ 100%) are found in the Cayman Sea, the northwest of the Colombia Basin, and close to the deep sills toward the northern Caribbean Sea boundary (Fig. 4c). Therefore, NADW enters the Caribbean Sea through the deeper sills in the Anegada Passage (1820 m) and the Windward Passage (1680 m) (Fig. 1). The spatial distribution of NADW concentration indicates that most of this water mass enters the Caribbean Sea through the Windward Passage and fills the entire deep Cayman Sea basin below 1500 m (Fig. 5e). It seems that NADW core crosses from the Cayman Basin to the western side of the Colombia Basin through the Jamaica Channel (1540 m) (Fig. 4c) as previously reported by Morrison and Nowlin (1982). Once in the Eastern Caribbean, it fills the deeper layers as it is the densest water mass in the study area (Fig. 3). Therefore, the NADW core is found close to the bottom but with a lower concentration due to mixing with AAIW, which has greater volume in the Eastern Caribbean than in the Cayman Sea (Fig. 5c). However, the NADW core concentration is > 80% in most of the Eastern Caribbean, except for relatively shallow areas such as the Granada Basin and the Central America Rise. Still, NADW concentration is > 70% below ~ 1500 m (Fig. 5e and f) as its upper boundary mixes with the shallower AAIW. NADW layer thickness increases with depth, reaching values > 4500 m (Fig. 4f and 5e). This NADW distribution makes the Cayman Sea waters below ~ 2000 m more ventilated for about 30 µmol kg−1 when compared to the Eastern Caribbean (Fig. Aux. 2d at 2000 m).

4.3 Water mass sub-annual variability in the Guajira region and relation to coastal upwelling

We evaluate sub-annual variations from intermediate and deep-water masses using in situ time series from three moorings close to the Guajira Peninsula (Fig. 1), in order to assess differences from the climatological perspective previously shown. Time series of nearly 11 months are available at 500, 700, 1200, and 1450 m. Conservative temperature (Θ) and absolute salinity (SA) at all depths do not largely differ from the Guajira node climatology (Fig. 6). However, sub-annual variations can be seen at all depths, but they reduce as depth increases (standard deviation and range in Table 2 and Fig. Aux. 4). The largest Θ and SA range occurs at 500 m in F3 (1.46 °C and 0.198 g kg−1). In all cases, Θ and SA variations seem to respond to mixture changes between the source water indices defined, as they maintain the linear relationship in the Θ–SA diagram (Fig. 6).

Fig. 6
figure 6

Conservative temperature (°C) and absolute salinity (g kg −1) diagrams from mooring daily time series at (a) F1, (b) F2, and (c) F3. Colors indicate the date (dd/mm/yy). Water source indices are shown in black circles: modified Antarctic Intermediate Water (mAAIW) and modified North Atlantic Deep Water (mNADW). WOA18 hydrographic data at Guajira (1300 m depth), as shown in Fig. 3, is indicated in black dots for reference; red dots indicate 500 m, 700 m, and 1200 m depths. Light gray lines show potential density anomalies (kg m−3)

Full size image
Table 2 Conservative temperature, absolute salinity, and water fraction (%) time series statistics from the mooring’s dataa
Full size table

Note that at 700 m depth, Θ and SA pairs are closer to the climatological values in F3, while salinity observations increase in F2 and F1 when compared to the WOA18 values. In all cases, observations indicate the presence of the AAIW core (lowest salinity below the depth of the salinity maximum). The main reason for these differences is that the WOA18 node is closer to F3 (Fig. 1). Besides, from F3 toward F1, the AAIW core increases its salinity as a consequence of mixing during its southwestern advection (Fig. Aux. 3). Therefore, the 11-month time series support the climatological results shown in Section 4.2 but showing large sub-annual variations due to changes in the water mass mixture.

In addition to the sub-annual variations, interannual variability can also affect the differences between observations and the climatological annual mean. There is evidence of intra-annual-to-multi-decadal variations in temperature and salinity of SUW, AAIW, and NADW (Johnson and Purkey 2009; Montes et al. 2016; Rhein et al. 2013). In addition to changes in the water mass properties, changes in their mixtures can affect density profiles in the Caribbean Sea (e.g., Andrade 2000), producing steric sea level variations. For example, steric trends above 800 m have been reported in the Caribbean Sea (Torres and Tsimplis 2013), with a basin mean of 0.6 ± 0.2 mm year–1 in the 1950–2010 period. These positive trends were mostly caused by thermosteric expansion (1.0 ± 0.2 mm year–1), partly compensated by negative halosteric trends (–0.3 ± 0.1 mm year–1). Decadal steric trends have large temporal and spatial variability in the Caribbean Sea, what is probably related to changes in the basin’s water masses.

To assess sub-annual changes of water mass fractions, we only have Θ and SA observations; therefore, only the mixing between two source water indices can be assessed simultaneously (Section 3). At 500 m, water mass results from SUW and AAIW mixing. Besides, at 1200 m and 1450 m, water mass results from AAIW and NADW mixing (Fig. 3). In these cases, accurate results are obtained from the OMP linear mixing model. Limitations occur at 700 m. At this depth, the water mass is dominated by AAIW but influenced by both SUW and NADW. We found that if mixing is assessed between AAIW and SUW, the dominant AAIW fraction temporal mean is larger (~ 95%), and its variability is strongly smoothed. If mixing is assessed between AAIW and NADW, the AAIW fraction temporal mean is ~ 76%, but its variability is coherent with temperature and salinity changes in the time series (Fig. Aux. 4). Therefore, we show these results (Fig. 7). We performed some tests including a constant value of dissolved oxygen using climatological data and modifying the weighting matrix, to perform the OMP analysis at 700 m depth using the three source water indices. From this assessment, we estimate that the AAIW fraction mean we report is reduced ~ 10% at this depth, due to not simultaneously including the three source water indices in the OMP model. However, temporal variations in AAIW fractions are not affected as a consequence of not including SUW in the OMP analysis.

Fig. 7
figure 7

Dominant water fractions at each depth of the daily time series, obtained from the OMP analysis using two water types. At 500 m AAIW (vs. SUW). At 700 AAIW (vs. NADW). At 1200 and 1450 m NADW (vs. AAIW). Moorings (a) F1 and F3 and (b) F2. (c) Schematic diagram showing the variation of water mass fractions due to upwelling intensification in the Guajira region (F2), when Ekman offshore transport enhances in response to wind stress intensification. (d) Low-frequency-normalized time series showing water fraction variability at depths presented in (c). The rotated wind stress (Tau U rot) is included. A rise in wind stress indicates an offshore intensification of the Ekman transport, enhancing coastal upwelling. Gray crosses indicate the days when zonal rotated wind stress was negative (shoreward), producing coastal convergence, what would force the opposite response of water mass fractions shown in (c)

Full size image

At 500 m, AAIW fraction (~ 88% in Table 2) dominates SUW, as at this depth, water mass is closer to the former source water index. AAIW is dominant at 700 m with fractions that reduce from F3 (80.7%) to F1 (76.0%), which is consistent with the AAIW westward advection in the Eastern Caribbean (Fig. 4b and Fig. Aux. 3). If 10% is added to this value (due to not including SUW in the model), results are in agreement with AAIW climatological concentration between 11° and 13°N in Fig. 5d. At 1200 and 1450 m, water masses result from the combination of AAIW and NADW; however, the latter dominates with 82% and 93% water fractions, respectively. As expected, the largest NADW concentration is found in the deeper time series. All water fractions time series show large sub-annual variability (Fig. 7a and b). AAIW fractions at 500 m variate with a range of 5.8 to 8.1% in the three moorings (Table 2). NADW range is 9.6% at 1200 m and 5.8% at 1450 m in F2. The water fraction range is larger in AAIW at 700 m (12.2 to 15.4%), probably influenced by mixing with SUW and NADW.

Most of the temperature, salinity, and water fraction residuals (after the mean and trend were removed) have large variability at periods between 91 to 111 days (Table 2). Therefore, these in situ time series have significant sub-annual variations at different frequencies, from which ~ 90 day’s period protrude. This period is consistent with reported mesoscale variability in the Caribbean Sea (e.g., Jouanno et al. 2008). The relation between changes in intermediate and deep-water mass fractions and mesoscale variability in the Caribbean Sea should be studied in more detail.

We further study the forcing behind the observed in situ time series sub-annual variability. Due to the mooring position and available timescales, we assessed the relation between these changes and the Guajira upwelling dynamics through the correlation between the wind stress of the rotated surface wind and water mass fraction temporal evolution.

The mean (standard deviation) of the low-frequency rotated zonal wind (Urot) is − 6.0 ms−1 (4.1 ms−1), − 7.0 ms−1 (3.5 ms−1), and − 8.0 ms−1 (2.3 ms−1) in F1, F2, and F3, respectively (Fig. Aux. 4d). Negative values indicate northern trade wind direction toward the west–southwest. The mean wind speed increases (and standard deviation reduces) toward the center of the Caribbean (F3), indicating the stronger presence of the Caribbean low-level jet. A similar pattern is observed in the wind stress with a minimum mean in F1 (0.10 Nm−2) and a maximum mean in F3 (0.13 Nm−2). Note that we reverse the wind stress sign so that positive values indicate coastal upwelling (Section 3.3). In this study, we assume that wind stress from the rotated zonal wind is a good indicator of coastal upwelling strength. Although this is a valid approach, it has its limitations. For example, it does not account for the effect of variations in the meridional component of the rotated wind nor take into account other factors that might affect the wind upwelling dynamics, such as variations in the vertical stratification (Section 4.1).

Wind stress drives flows in the Ekman layer directly to 100 to 200 m depth and geostrophic flows to depths of 1000 to 2000 m, associated with divergence of Ekman flows and pressure gradients which they cause (Pond and Pickard 1983). This seems to be the case in the Guajira upwelling region due to denser surface water (Fig. 2) and its reported signature in sea level (Torres and Tsimplis 2012). Therefore, we investigate the depth at which the coastal divergence is able to force an upward flow, across the continental margin in the Guajira region.

To achieve this purpose, we assess the temporal evolution of the wind stress and its impact in the coastal upwelling system (Fig. 7c and d). When the wind stress increases, it enhances offshore Ekman transport and coastal upwelling. This causes coastal sea level to drop and induced geostrophic currents force deeper water upward. Therefore, at 500 m, the AAIW fraction is expected to rise, as the AAIW core is located below this depth (~ 700 m). Similarly, at 700 m, 1200 m, and 1450 m, the NADW fraction is expected to rise, as the NADW core is placed toward the bottom of the basin. On the contrary, a decrease in the wind stress will weaken the offshore Ekman transport and coastal upwelling, reducing all water mass fractions previously mentioned. Therefore, coastal upwelling (rotated wind stress) is expected to positively correlate with water mass fractions shown in Fig. 7d.

At each mooring, all the water fraction time series show positive correlations among them. The smaller correlation is between AAIW@500 and NADW@700 in F3 (0.35). All other correlations are > 0.50. The largest correlation occurs in F2 (Fig. 7d), between NADW@1200 and NADW@1450 (0.87). These results indicate the correspondence of water fraction variations at different depths at all the three moorings, suggesting the presence of a similar forcing able to affect water fractions at all depths, such as geostrophic flows forced by coastal divergence.

In F1, there is only a significant correlation (0.25) between the rotated wind stress and NADW@700. In F3, there are significant correlations between the rotated wind stress and AAIW@500 (0.36) and NADW@700 (0.41). At these moorings, the correlation is stronger at a depth of 700 m. Bear in mind that at these moorings, the bottom is 750 m depth, close to the AAIW core. Besides, due to their position (Fig. 1), these moorings are not placed in the main upwelling area. Furthermore, F1 is close to the Magdalena River outflow, probably affecting upper vertical stratification and reducing surface density, horizontal pressure gradients, and geostrophic induced flow.

F2 is the best mooring to assess the Guajira upwelling dynamics due to its position, depth of installation (1500 m), and number of recording instruments (4 CTDs). The rotated wind stress correlates positively with AAIW@500 (0.26) and NADW@700 (0.48). On the contrary, negative correlations appear with NADW@1200 (− 0.24) and NADW@1450 (− 0.13). We further inspected the relation between time series during specific periods. For example, in June to July, the rise and drop of the rotated wind stress indicate the effect of the mid-summer drought (Section 2), which affects all water mass fractions (Fig. 7d). Also at the end of October to the beginning of November, the upwelling relaxation due to onshore wind stress (gray crosses in the rotated wind time series) forces a reduction in all the water fractions. Similar responses of the water fractions are found in F1 and F3 as a consequence of strong changes in the rotated wind stress (not shown).

These results suggest that coastal divergence in the Guajira region is able to affect regularly water mass fractions down to 700 m depth. However, in all moorings, this relationship weakens at 500 m when compared to 700 m, where correlations with wind stress are stronger. This effect could be explained by a more efficient induced geostrophic currents dragging NADW up the continental margin to 700 m; however, this would not explain the better correlation at 700 m depth in the deeper F2. We suggest that eddy-related mesoscale variability found in the time series residuals (Table 2) produces a stronger effect at 500 m depth (weakening the correlations with wind stress), as their kinetic energy reduces with depth (e.g., Jouanno et al. 2008).

In F2, where we have time series at 1200 and 1450 m depth, the upwelling seems not to regularly affect these water mass fractions. However, during periods of strong changes in wind stress forcing, upwelling is able to affect the water fractions at these depths. We also assessed lagged correlations to seek for delays between wind forcing and changes in the water fractions, as well as delays between changes of shallower water fractions and deeper ones. In general, results did not significantly improve, what indicates that upwelling dynamics affect the water column with little delay (hours to few days).

5 Summary and final remarks

Our first objective is to assess the CSW seasonal variability in the Caribbean Sea from a climatological approach, studying its properties at the mixed layer depth (MLD). Air–sea heat fluxes, river runoff, and wind seasonality modulate the CSW properties at the MLD. Particular CSW properties are found in the Southern Caribbean Upwelling System, the Gulf of Mosquito-Darien sub-basin (dilution sub-basin), and the Venezuela and Granada basins (Orinoco outflow) (Fig. 2), what agrees with literature (e.g., Beier et al. 2017; Montoya-Sánchez et al. 2018a). Due to the low-density surface waters (warm and fresh), subsurface or deeper waters are not formed in the Caribbean Sea at any time of the year.

SUW enters the Caribbean Sea through the northeast Caribbean boundary at ~ 150 m depth and is advected westward by the Caribbean Current, reducing its salinity by mixing. SUW core is deeper north of ~ 15°N (Fig. 2j) as it founds a deeper mixed layer (Fig. 2a), while it is shallower toward the southern side of the Eastern Caribbean, where the mixed layer is also shallower. These spatial variations are due to wind stress curl forced by the Caribbean low-level jet (Montoya-Sánchez et al. 2018a).

In all the study area, CSW at the MLD is fresher than the SUW core and with larger seasonal salinity variability (Fig. 2). The larger the vertical salinity differences, the stronger the halocline and upper-density stratification, which forms a barrier layer that reduces vertical mixing. A barrier layer occurs when the density upper mixed layer is shallower than the isothermal layer (Sprintall and Tomczak 1992), as is the case in the Caribbean Sea (Fig. 2a and Fig. Aux. 1). Besides, in this basin, the barrier layer produces spatial and seasonal variations in the static stability above the maximum salinity layer, in response to the CSW and SUW changes (Fig. 2l, m, and o).

Note that the static stability variations will have important effects on the Caribbean Sea dynamics, including coastal upwelling, mixed layer temperature, and mesoscale variability. For example, in the Caribbean Sea, surface diluted waters coincide with warmer waters, probably due to the strong vertical static stability, which limits the absorption of solar radiation to a thin surface layer (Ruiz-Ochoa et al. 2012) and reduces the inflow of colder subsurface water into the mixed layer (e.g., Ke-xin and Fei 2021). Furthermore, in the Caribbean Sea, anticyclonic eddies dominate mesoscale variability, which intensify on their westward path. Horizontal density gradients related to upwelling and the Orinoco river plume have been proposed as the cause (van der Boog et al. 2019). We show that upwelling and river plumes also affect vertical static stability between the CSW and SUW core in the Eastern Caribbean. Therefore, we suggest that the static stability spatial variations might affect eddy dynamics in this basin. A reduction of the static stability from the Lesser Antilles toward the northwest would facilitate eddy kinetic energy increase in this direction, allowing the strengthening of eddies (Fig. 2l).

In this study, we also present the individual volume and spatial distribution of subsurface, intermediate, and deep-water masses in the Caribbean Sea from a climatological perspective. To perform the OMP analysis, we first define source water indices of these water masses when entering the Caribbean Sea. This assessment is done in terms of TEOS-10 (Table 1). The mixture of four water types can describe the Caribbean Sea hydrography (Fig. 3). The Caribbean Surface Water above the mixed layer depth, the North Atlantic Subtropical Underwater at ~ 150 m depth which is denoted by its salinity maximum, the Antarctic Intermediate Water with its core ~ 700 m depth and characteristic minimum salinity and dissolved oxygen, and the North Atlantic Deep Water which is the densest water in the basin.

The OMP analysis applied below the salinity maximum indicates that 46% of the Caribbean Sea volume corresponds to NADW, 38% to AAIW, and 16% to SUW. SUW fraction distribution is consistent with the behavior described from the salinity maximum (Fig. 2j). AAIW volume in the Eastern Caribbean is higher than in the Cayman Sea, and the position of its core coincides with the spatial distribution of the minimum salinity in the water column (Fig. Aux. 3), what shows a small seasonality. Most NADW enters the Caribbean Sea through the Windward Passage and fills the Cayman Sea below ~ 1000 m (Fig. 4 and 5). In the Eastern Caribbean, NADW enters through the Anegada Passage and the Jamaica Channel coming from the Cayman Sea, with higher fractions placed toward the bottom of the basin.

The assessment of the most important water mass fractions in the Caribbean Sea allows us to observe their distribution in the different basins, as well as to follow their pathways after entering the Caribbean Sea. This information is useful to assess heat, oxygen, and nutrient transport, among other properties with implications on the biosphere. Besides, this information serves as a starting point to assess the relation between water mass variability and steric sea level, which is a main component of the observed sea level rise (Masson-Delmotte et al. 2021). This has become a priority topic as the planet continues its warming trend, putting at risk the low-laying coasts in the Caribbean Sea. In particular, steric sea level trends have been observed in the Caribbean Sea (Torres and Tsimplis 2013) and are expected to continue during the entire century due to global warming (Bustos Usta and Torres Parra 2021).

Conservative temperature and absolute salinity from eight in situ time series in the Guajira upwelling region agree with the climatological behavior (Fig. 6). However, large sub-annual variations in water properties are detectable in all the time series and decrease with depth (Table 2). Besides, interannual variations in water masses mixing and properties might influence differences between observations and the annual climatological mean.

At 500 and 700 m, AAIW is the dominant water type, while at 1200 and 1450 m, NADW is dominant. Water mass fractions at all depths in the same mooring significantly correlate, what suggests a similar forcing such as geostrophic flows associated to coastal upwelling. We show a temporal relation between coastal upwelling strength (rotated wind stress) and water mass fractions. Upwelling dynamics influence water mass fractions regularly down to 700 m depth in the Guajira continental margin; however, during periods of strong changes in wind stress, water fraction relation can be observed down to 1450 m depth. This mechanism seems to affect the water column in periods no longer than few days.

Correlations between wind stress and water fractions at 500 m are weaker than at 700 m depth probably due to strong mesoscale activity in the region (Jouanno and Sheinbaum 2013). Observed sub-annual variations in the Caribbean Sea water fraction peak close to 90-day period (Table 2), indicating a relation with mesoscale variability in the Caribbean Sea, which has large energy at this period (Jouanno et al. 2008). This variability (e.g., shrinking–stretching of the water column due to eddies) seems to produce larger effects at 500 m than deeper, masking the effects of coastal upwelling at this depth. Besides, other mechanisms such as vertical stratification (Section 4.1) would act simultaneously with the upwelling dynamics and mesoscale variability, affecting the temporal evolution observed in water mass fractions.

Reported findings on the water mass distribution and variability can have an impact on the biosphere (Puerta et al. 2020). For example, spatial and seasonal variability of surface waters and static stability would affect the vertical mixing and fertilization of surface waters (Lozier et al. 2011). Besides, the distribution of the AAIW and the NADW might affect ecosystems due to differences in water properties such as dissolved oxygen (e.g., Keeling et al. 2010; Stendardo et al. 2015). AAIW is distinguished by its mid-depth oxygen minimum and maximum silicate that is an important nutrient for specific biota, while NADW is more ventilated than AAIW. Note that in the Caribbean Sea, the oxygen content of NADW is the nearly double the oxygen content of AAIW (Table 1). Therefore, the distribution and properties of water masses will have an effect on biodiversity and biogeography in the Caribbean Sea at different depths. Besides, water mass distribution have an effect on the water column density (steric changes), which affects the mean sea level rise (Wunsch et al. 2007). Therefore, improving our understanding of the temporal variation of the water mass properties and distribution in the Caribbean Sea, including those caused by global climate change, will provide important information to assess future mean sea level variations and threats to ecosystems diversity in the basin.