Freshwater exchanges and surface salinity in the Colombian basin, Caribbean Sea

Despite the heavy regional rainfall and considerable discharge of many rivers into the Colombian Basin, there have been few detailed studies about the dilution of Caribbean Surface Water and the variability of salinity in the southwestern Caribbean. An analysis of the precipitation, evaporation and runoff in relation to the climate variability demonstrates that although the salt balance in the Colombian Basin overall is in equilibrium, the area south of 12°N is an important dilution sub-basin. In the southwest of the basin, in the region of the Panama-Colombia Gyre, Caribbean Sea Water is diluted by precipitation and runoff year round, while in the northeast, off La Guajira, its salinity increases from December to May by upwelling. At the interannual scale, continental runoff is related to El Niño Southern Oscillation, and precipitation and evaporation south of 12°N are related to the Caribbean Low Level Jet. During El Niño years the maximum salinification occurs in the dry season (December-February) while in La Niña years the maximum dilution (or freshening), reaching La Guajira Coastal Zone, occurs in the wet season (September-November).


Introduction
The Caribbean Sea is considered a concentration basin (i.e. evaporation exceeds precipitation) even though the increase of salinity is not enough to cause deep vertical convection [1][2][3]. Schmidt et al. [4] concluded that Caribbean salinity is not significantly affected by freshwater runoff and primarily reflects the evaporation/precipitation ratio. On the other hand, Yoo and Carton [2] noted that the river discharge and the annual average rate of advection of freshwater into the Caribbean Sea from the Atlantic differ by just over 10%.
The transport of water through the succession of Grenada, Venezuelan, Colombian, Cayman and Yucatan basins that constitute the Caribbean forms part of the western boundary current system of the North Atlantic Ocean and therefore is associated with the global meridional overturning circulation. Surface currents in the upper 50-to-100 m layer transport Caribbean Surface Water (CSW), which is formed by dilution of Atlantic Water by freshwater PLOS

Materials and methods
Precipitation over the basin was obtained from the monthly long-term mean data of the Climate Prediction Center Merged Analysis of Precipitation (http://www.esrl.noaa.gov/psd/data/ gridded/data.cmap.html) with a resolution of 2.5˚x 2.5˚ [22]. The data of Skin Temperature from the NCEP/NCAR Reanalysis with a spatial resolution of 1.86˚x 1.86˚(http://www.esrl. noaa.gov/psd/data/gridded/data.ncep.reanalysis.derived.surfaceflux.html) [23] and Net Latent Heat Flux from North American Regional Reanalysis, 0.25˚x 0.25˚(ftp://ftp.cdc.noaa.gov/ Datasets/NARR/Monthlies/monolevel/apcp.mon.mean.nc) [23]. Evaporation during the common period of the three series (January 1979 to December 2012) was computed using the method described by Gill [24]: where E is the evaporation rate (m s -1 ), Q e is the latent heat flux (W m -2 ), ρ is the density of water (kg m -3 ), L v = 2.5008x10 6 -(2.3x10 3 T) is the latent heat of evaporation (J kg -1 ), and T is the surface temperature (˚C).
NCEP/NCAR and NARR reanalyses are among the most dependable, have a reasonably high resolution, and are widely used by the geophysical community. As a check on data quality, our evaporation series estimated from NCEP/NCAR data at (10.625˚N, 79.375˚W) was compared to that in the ERA-Interim reanalysis (http://www.ecmwf.int/en/research/climatereanalysis/era-interim) for a location in the Caribbean representative of the area south of 12( 10.75˚N, 78.25˚W). The two series had similar characteristics over the period of comparison 1979-2014, with high correlation r = 0.93 and p-value = 0.001, which suggests good agreement between the two series. The NCEP/NCAR estimates are used for consistency with the other analyses in this paper.
Data from 36 rivers discharging into the Colombian Basin (Table 1) [25] and the River Discharge Database Version 2.0 for Nicaragua (http://nelson.wisc.edu/sage/data-and-models/riverdata/index.php). Data in Table 1 were compared and completed with data of river discharges to the Colombian  Basin including data from Restrepo et al. [18], Restrepo and Kjerfve [20] and Milliman and Farnsworth [26]. There are no reports of the estimation error of discharges. The data are separated into four groups in the table and figures for convenience of later reference. High-resolution (0.25˚x 0.25˚) gridded ocean climatology was obtained from the National Oceanographic Data Center Web site (http://www.nodc.noaa.gov/OC5/woa13/woa13data. html), hereafter WOA13v2 data [27][28]. Among the few world databases of observed hydrography, this has the best resolution and is the most widely used. WOA13v2 hydrographic data provide objectively analyzed climatological mean fields, with greater vertical and spatial resolution than previously available [27][28][29].
Climatology of temperature was derived from 1 to 1 32 casts per quadrant (grid square) and salinity from 1 to 37 casts. Though salinity observations in WOA13v2 are fewer than those of temperature, it is the best available source of information on in situ salinity. Observations were available for between 1 and 11 months of the year in each quadrant.
Monthly mean Sea Surface Salinity (SSS) from Aquarius/SAC-D observatory was downloaded with spatial resolution of 1˚x1˚ [30][31] from the Physical Oceanography Distributed Active Archive Center (PODAAC, http://podaac.jpl.nasa.gov/aquarius). We used Combined Active-Passive (CAPv3) Algorithm for the sea surface roughness correction to enable the retrieval of SSS, from September 2011 to April 2015 [31]. To compare the salinity data from WOA13v2 and Aquarius/SAC-D we linearly interpolated the 1-degree Aquarius data onto a 0.25-degree grid.
The climate of Colombia has been described by Mesa et al. [32]. The major dry season from December to February (DJF) is followed by a minor wet season from March to May (MAM), minor dry season ("midsummer drought") from June to August (JJA), and major wet season from September to November (SON). We used the same seasons in our analysis. Seasonal total discharge in the rivers was calculated by regions (Table 1) Table 1, river discharge in the Colombian Basin occurs mainly south of 12˚N and can be represented by two rivers (Atrato and Magdalena) which combined contribute over 60% of the total runoff. Both rivers have long records and were used for further analysis.
For each forcing agent of SSS (evaporation, precipitation, and runoff) we have performed a seasonal fitting using Least Squares Fitting and estimations of uncertainties [33][34][35]: where A 0 is the mean value, A α and A s are the annual and semiannual amplitudes for each time series; ω = 2π/365.25 is the annual radian frequency; φ α and φ s are the phases of annual and semiannual harmonics referred to the beginning of the year; and t is the time. The term F nonseason represents the anomalies and contains the non-seasonal variability (interannual and mesoscale). It was used to understand the interannual freshwater input into the basin, through comparisons with climatic indexes. Of the different El Niño Southern Oscillation (ENSO) indexes, the Oceanic El Niño Index appeared the most representative of the salinity forcing agents and is the one used here (ONI, http://www.cpc.ncep.noaa.gov/products/analysis_monitoring/ensostuff/ ensoyears.shtml). We used the Caribbean Low Level Jet Index (CLLJ Index, Wang, 2007), defined by taking the negative of the 925hPa zonal wind anomalies in the region of 12.5˚-17.5˚N, 80 o -70 o W. Since the lower tropospheric winds in the region are easterly, the definition indicates that when the CLLJ Index is positive, the CLLJ is anomalously strong.
All time series used in this work are long enough (>25 years) to represent properly the interannual variability. To account for the effects of autocorrelation within each time series, the Effective Degrees of Freedom (N ef ) were calculated following Trenberth [36].
To further investigate the forcing of interannual freshwater inputs in the Colombian Basin on water masses, we used information by seasons (three-monthly periods) of modeled SSS and currents near the coast during typical El Niño and La Niña years. These results correspond to outputs for the Colombian Basin over the period from 1979 to 1998 of a global run of the Parallel Oceanic Circulation Model (POCM-4C, courtesy of Dr. Robin Tokmakian from the Naval Postgraduate School, Monterey, CA, USA). The spatial and temporal resolution of the model was 0.25˚x0.25˚and one month, respectively. The POCM model [37][38] reproduces properly in qualitative terms the seasonal and interannual variability of large scale circulation [39][40].

Results and interpretation Fresh water inputs
The climatological seasonal maps of evaporation minus precipitation (E-P) (mm month -1 ) are shown in  During the "midsummer drought" (JJA) the lowest values of E-P occur close to Central America, and during the main wet season (SON) the area of dilution extends as far as 76˚W. The spatial pattern of dilution and concentration basins during the second half of the year reflects the Walker-type convection cell and the divergence of winds over the Colombian Basin [41][42][43].
The total annual mean river discharge into the Colombian Basin is estimated at 15397 m 3 s -1 (Table 1). This is equivalent to about~50% of the annual mean discharge of the Orinoco River (31000 m 3 s -1 ) [7] into the eastern Caribbean. Over 60% of total water discharge directly into the Colombian basin arises from only two rivers, the Magdalena (7100 m 3 s -1 ) and the Atrato (2400 m 3 s -1 ). Rivers north of 12˚N make a smaller contribution to the overall budgets. The climatological monthly mean of rivers draining to the Colombian Basin has a strongly seasonal signal, as can be seen in Taking into account the E-P and runoff shown in Figs 2 and 3, most of the surface water dilution in the Colombian Basin occurs south of 12˚N. Climatological monthly means of evaporation (E), precipitation (P) and runoff (R), as well as the resulting budget between them, E-(P+R), are depicted in Fig 4A and 4C for the entire Colombian Basin and for the area south of 12˚N Fig 4B, 4A and 4D, respectively.  Table 1 and Fig 1. The thin lines represents the seasonal fitting, and the respective percentage of explained variance is included.
Overall, in the Colombian Basin, precipitation and evaporation are up to 5 times higher than runoff (Fig 4A) while south of 12˚N all are more similar in magnitude (Fig 4B). From May to November, the Colombian Basin is a dilution basin while from December to April it is a concentration basin (Fig 4C). The region south of 12˚N suffers net dilution from April to December and only weak concentration in the remaining months ( Fig 4D). Total evaporation in the Colombian basin is 57965 m 3 s -1 and total precipitation 49184 m 3 s -1 .
The net evaporation for the Colombian Basin E-(P+R) is -6616 m 3 s -1 , an order of magnitude smaller than E, P or R. But the freshwater budget for the area south of 12˚N is determined by evaporation of 12959 m 3 s -1 , precipitation of 17105 m 3 s -1 and runoff at 11845 m 3 s -1 , and so the net balance of E-(P+R) = -15991 m 3 s -1 . Given that, the sum and its constituent terms all have the same order of magnitude; this sub-region appears to be a zone of net dilution. However, relative errors in each of the terms may be large, and more observations are needed to confirm this result.

Surface water properties
Using the WOA13v2 data in the Colombian Basin, the main Caribbean water masses reported in previous works [5,15,[44][45][46] can be identified. The mean potential temperature and salinity profiles as a function of depth, averaged over the whole Colombian Basin with their respective minimums and maximums (Fig 5), show a wide variation of the salinity of the CSW. Centered at 150 m, the Subtropical Underwater (SUW) is seen as a typical salinity maximum of 36.5 above the deeper Tropical Atlantic Central Water (TACW) and Antarctic Intermediate Water (AAIW) layers.
The spatial and seasonal changes can be seen in detail in the θ-S diagrams of surface waters down to 100 m depth (Fig 6) averaged by depths and regions as follows: north of 12˚N; south of 12˚N and east of 74˚W (La Guajira upwelling zone); and south of 12˚N and west of 74˚W (the intense precipitation area of the Panama-Colombia Gyre). Surface waters are within the intervals reported for CSW (temperature between 25 and 29˚C; salinity higher than 35.1) but, in the area of the Panama-Colombia region south of 12˚N, waters above 30 m are fresher and warmer all year than those north of 12˚N. Meanwhile, in La Guajira region, surface waters are significantly colder than in the other regions from December through May because of strong upwelling [47][48]. During these months, a thin (~10 m) low salinity surface layer is present, but waters between 10 and 100 m are significantly saltier than in the other regions (also because of the upwelling). In JJA, surface water in la Guajira is colder but with the same salinity as the Panama-Colombia region (lower than the Central Caribbean). During SON, when upwelling is weaker, water masses of the three regions become similar, except for the shallowest surface waters (above 5 m) of Panama-Colombia and La Guajira whose salinity remains low.
The seasonal distributions of WOA13v2 SST (Fig 7) show the strong surface cooling produced by the upwelling in La Guajira Coastal Zone particularly during DJF and MAM. During JJA and SON, although La Guajira region remains locally cooler, surface heating is evident throughout the basin, including the regions of anomalously high temperatures in the Darien and Mosquitos Gulfs (see Fig 1). These anomalies were reported and correlated with surface forcing by Ruiz-Ochoa et al. [41] based on Pathfinder satellite SST. The strong similarity of the temporal and spatial patterns from WOA13v2 and Pathfinder SST confirms the capability of the former to characterize the climatology of hydrography in the basin, although there are fewer salinity than temperature observations in the database.
On the basis of WOA13v2 salinity surface quarterlies (Fig 8), the central basin salinities are generally highest in MAM and JJA, and lowest in SON, the period of most rain and advection of waters diluted by the Orinoco from the Atlantic [5,11]. The cooler waters of La Guajira region are also generally saltier than the rest of the basin. In contrast with the saltier waters off La Guajira, fresher waters of salinities as low as 34.2 occur in the Darien and Mosquito Gulfs, coincident with higher temperatures. These low salinity waters extends eastwards along the Colombian coast in JJA reaching their maximum extension in SON, when precipitation and runoff are maximum. The eastward extension of low salinity waters is minimal in MAM.
The spatial distributions of sea surface salinity and temperature suggest that in the Colombian Basin the CSW is diluted south of 12˚N, where it acquires lower salinities and higher temperatures. Waters in the basin are more homogenous in MAM, after the main dry season, when they are saltier, and in SON, the main rainy season, when they are fresher. Off La

Salinity exchanges Colombian basin
Guajira, the colder and saltier water associated with the upwelling of SUW in DJF and MAM becomes fresher and warmer in SON because of the advection of waters from the west. From DJF through MAM, the intensity of the CLLJ and wind stress curl along the La Guajira Coast is most favorable to upwelling [41].
The recent availability of remotely sensed sea surface salinity from the Aquarius mission provides an independent check on the salinity field for the basin. Although Aquarius data are currently available only from September 2011 to April 2015 and are known to be unreliable close to shore, they still provide useful information on the region. We calculated seasonal means of these data using only pixels flagged as valid and free from land contamination. The seasonal composites over this period (2011)(2012)(2013)(2014)(2015) showed higher salinity values (between 0.25 and 0.5) but a similar seasonal pattern to the WOA13v2 data (r = 0.72, p-value = 0.001) despite the different record lengths and the low spatial resolution of Aquarius.

Interannual variability of freshwater inputs and water masses
The seasonal cycles explain 61% of the total variance of E, 80% of P, and 53% and 30% of river outflow (R) from Magdalena and Atrato rivers, respectively. Fitting errors of amplitudes and phases are low. Monthly means of Magdalena runoff, M nonseason , from 1950 to 2004, 648 observations, correlate with the ONI Index with an r = -0.52, p-value<10 −10 , N ef = 168 and a lag of one month. This suggests that during El Niño (La Niña) events, the Colombian Basin receives less (more) runoff. Furthermore, the Magdalena anomalies of discharge (M nonseason ) are correlated with the Niño1+2 Index, which represents the SST in the extreme Eastern Tropical Pacific, with r = -0.64, p-value<10 −11 , N ef = 172 and a lag of one month. Similar values of correlation coefficient and statistical significance were found between the Atrato river outflow and ONI index. shows the relation with the interannual variability through the CLLJ Index (Fig 9A and 9B), and ONI (Fig 9C). The sum of P+R in Fig 9B and 9C is one order of magnitude higher than E. The figure shows that E and R decrease during El Niño, and increase during La Niña periods. The pattern for precipitation in relation with El Niño or La Niña is not clear, whenever it is not well correlated with ENSO.
Output from the POCM-4C was used to test qualitatively if ENSO phases affect the river discharges sufficiently to modify the coastal surface water masses. Here Caribbean SSS maps were extracted from the global run over the period 1980 to 1998 as explained in the Methods section. Results for contrasting years show that during El Niño of 1983 (Fig 10) the low salinity coastal water in the Panama-Colombia Gyre was less extended and the high salinity off La Guajira was more extended than in the climatology of POCM, for both dry and wet seasons (Fig 11). Conversely, during La Niña of 1988 (Fig 10), the low salinity coastal water was more extended than in the climatology of POCM (Fig 11) and extended as far as La Guajira. Although the model captures the seasonality only qualitatively, it represents the general oceanography of the Colombian Basin and the spatial distribution of the water masses. Therefore, it serves as a first approximation to their interannual behaviour and represents the ENSO variability reasonably well.

Discussion and conclusions
The Caribbean Sea is generally considered a concentration basin where evaporation exceeds precipitation [1][2][3][4]. Remotely sensed SSS and numerical modelling indicates that, on an interannual basis, the salinity of water entering the Caribbean through the Lesser Antilles can vary by 0.5 psu [8]). This results from variability of the Amazon plume caused by changes in discharge rates and in the ocean and atmospheric forcing that govern its extent. Thus interannual anomalies of salinity unrelated to local conditions propagate westward through the central Caribbean and on into the Gulf Stream system. On the other hand, our results show that variability in the southwestern Caribbean is related to the local conditions. Unlike the rest of the Caribbean, the Colombian Basin is not a concentration basin, and south of 12˚N it is a strong dilution basin in which the contribution of runoff to the water balance is as important as evaporation and precipitation (Fig 5D).
The net dilution in the southwest and upwelling in the northeast of the southern Colombian Basin modifies CSW, which shows wide variation of its salinity, as noted by Andrade and Barton [49]. Furthermore, Corredor [47], reported alternation of surface waters off La Guajira Peninsula between properties of the Panama-Colombia Countercurrent and wind-induced coastal upwelling. In situ observations by Andrade and Barton [48] and simulations by Lonin et al. [50] found that warm and low salinity waters from the Panama-Colombia Gyre block the southwestward extension of La Guajira coastal upwelling that normally reaches 10.5˚N. During upwelling, we found a thin surface layer (0-10 m) of low-salinity water with very salty water between 10 and 100 m. Presumably when upwelling is weaker off La Guajira, the western water mass penetrates further to the east.
Interannual variability of the evaporation and precipitation in the south of the Colombia Basin correlates better with the local forcing of the CLLJ Index than with ENSO (ONI). This permanent low-level wind jet over the basin [51][52][53][54][55] enhances evaporation, transports humidity and forms strong convection over topography in Central America. Conversely, runoff is better correlated with ENSO (ONI) because it originates principally in continental Colombia where, during El Niño events, precipitation and runoff decrease, while La Niña events have the contrary effects [19,32,[56][57][58][59][60]. Even though the annual cycles of precipitation and runoff are related, the deseasonalized time series are poorly correlated (r = 0.19, p-value < 0.03, N eff = 129 over 288 observations), because of the isolation of the catchment area of the Colombian rivers in the northern Andes from the Caribbean. Although both spatial patterns of precipitation over the Colombian mainland and their temporal variability are complicated, it has been recognized that ENSO is the greatest single cause of interannual variability within the region (e.g. Poveda et al. [57]) as indicated here by the significant correlation between the ONI and Magdalena River outflow.
Amador [54] reported that during El Niño (warm) phases, stronger than normal CLLJ winds occur, while during La Niña (cool) phases the CLLJ is weaker, in both cases especially during summer. A stronger CLLJ was found to be associated with greater than normal precipitation in the western Caribbean and weaker precipitation in the central Caribbean, while a weaker CLLJ produced the opposite. Our results for that portion of the Colombian basin south of 12˚N show that precipitation is indeed positively correlated with the CLLJ Index, as expected. However, the precipitation anomalies were not well correlated with ENSO. We found that during the El Niño dry season (boreal summer) lower runoff results in less dilution of CSW in the Panama-Colombia Gyre and greater presence of upwelling waters (enhanced also by stronger winds of the CLLJ). In contrast, during the wet season of La Niña, runoff extends dilution beyond its normal range to reach as far as La Guajira Coastal Zone.
The interplay between warm and low salinity waters from the Panama-Colombia Gyre and cold and high salinity waters from La Guajira is related to the zonal wind circulation that plays a role in the Caribbean-Pacific connection [41][42][43]. The CLLJ is the surface component of the cell with upward convective processes at the Caribbean side of Central America, and low-level divergence and downward drying over La Guajira. According to Andrade and Barton [42] and Hidalgo et al. [43], the convergent zone is connected with another cell in the west, with the sinking of dry air over the Eastern Tropical Pacific, where the drying favors the southward displacement of the Inter Tropical Convergence Zone (ITCZ). This could be the link between ENSO and CLLJ that produces the seasonal and interannual patterns of water masses observed here.
La Guajira upwelling is part of the South Caribbean Upwelling System [61]. It is characterized by strong physical forcing [41,48,[62][63], strong mixing of subsurface water with surface water, but low chlorophyll concentration [63][64]. Different explanations reported for the low productivity in La Guajira include oligotrophic source waters [16], the upwelling of Panama-Colombia undercurrent water [63], a greater depth of the SUW in front of La Guajira than off Venezuela, or reduced efficiency of wind energy transfer [64]. Our findings indicate that from June to November, surface waters off La Guajira (see Figs 6 and 8) are similar to the western diluted CSW; especially during La Niña years (see Fig 10). This low salinity water, originating in the southwest of the basin, favors a strong surface stratification that inhibits vertical mixing and enhances surface heating and thus, plays a role in the low productivity off La Guajira.
Based on observed surface temperature, salinity and runoff, a first evaluation of the seasonal surface water balance of the Colombian Basin has been made. Results demonstrate that south of 12˚N in the Colombian Basin, the surface water balance indicates a net dilution, responsible for producing a strong dilution of CSW. This water reaches the La Guajira region from June to November, where it dilutes saltier water produced by local upwelling.
At interannual scale, precipitation and evaporation are related to variability of the CLLJ, while runoff is related to ENSO. In turn, the CLLJ can change in relation to ENSO. Since south of 12˚N runoff has the same order of magnitude as precipitation, the result is an anomalous decrease in the area of dilution and persistence of saltier water off La Guajira during the dry season of El Niño. In contrast, during the wet season of La Niña, the area of dilution extends beyond its normal range to reach La Guajira Coastal Zone. The intensified seasonal invasion of La Guajira upwelling by warm and freshwater from the Panama-Colombia Gyre during La Niña events favors surface stratification and can contribute to the low productivity of this upwelling system. Supporting information S1 Appendix. List of acronyms. (DOCX)