Remote sensing of salinity using satellite-mounted microwave radiometers provides new perspectives for studying ocean dynamics and the global hydrological cycle. Calibration and validation of these measurements is challenging because satellite and in situ methods measure salinity differently. Microwave radiometers measure the salinity in the top few centimeters of the ocean, whereas most in situ observations are reported below a depth of a few meters. Additionally, satellites measure salinity as a spatial average over an area of about 100 × 100 km2. In contrast, in situ sensors provide pointwise measurements at the location of the sensor. Thus, the presence of vertical gradients in, and horizontal variability of, sea surface salinity complicates comparison of satellite and in situ measurements. This paper synthesizes present knowledge of the magnitude and the processes that contribute to the formation and evolution of vertical and horizontal variability in near-surface salinity. Rainfall, freshwater plumes, and evaporation can generate vertical gradients of salinity, and in some cases these gradients can be large enough to affect validation of satellite measurements. Similarly, mesoscale to submesoscale processes can lead to horizontal variability that can also affect comparisons of satellite data to in situ data. Comparisons between satellite and in situ salinity measurements must take into account both vertical stratification and horizontal variability.
A synthesis of present knowledge about the formation and evolution of vertical and horizontal variability in near-surface salinity at scales relevant to satellite salinity is presented.
L-band microwave radiometers on both the Soil Moisture Ocean Salinity (SMOS; Mecklenburg et al. 2012) and Aquarius/Satélite de Aplicaciones Científicas-D (SAC-D) (Lagerloef 2012) satellites have now demonstrated that they are capable of measuring sea surface salinity (SSS). They provide near-global coverage, a spatial resolution ranging from 43 to 150 km, and a precision useful for detailed oceanographic studies, that is, ±0.2 practical salinity scale (pss) [salinity is a dimensionless quantity and will be reported on the Practical Salinity Scale of 1978 (PSS-78; IOC et al. 2010, and references therein) in the rest of the text] on monthly time scales and 100 × 100 km2 spatial scales (Drucker and Riser 2014; Hernandez et al. 2014; Hasson et al. 2013). This new capability provides an unprecedented global view of surface salinity, a key state variable that determines ocean circulation and is tied to the global water cycle (Reul et al. 2014c). These satellite-derived salinity data provide new insight into the spatial and temporal variability of SSS (Alory et al. 2012; Busecke et al. 2014; Hasson et al. 2014; Kolodziejczyk et al. 2015; Lee et al. 2014; Menezes et al. 2014; Qu et al. 2014; Reul et al. 2014a).
The success of satellite salinity measurements suggests new possibilities of using global maps of salinity to monitor and understand ocean dynamics and the global hydrological cycle. However, calibration and validation of satellite-retrieved salinity is an ongoing process that requires comparison of satellite SSS values with spatially and temporally collocated in situ values. There are two key differences between satellite and in situ salinity. First, because of the short penetration depth of microwave radiation into the ocean (Swift 1980), microwave radiometers measure salinity in the top few centimeters of the ocean. In contrast, in situ measurements commonly used for calibration and validation (e.g., Argo floats, moorings, and ship observations) are made at depths of a few meters (Fig. 1). Second, a satellite measures salinity as a spatial average over the satellite’s footprint, whereas in situ sensors provide data at a single point [SMOS synthetic antennas have variable elliptical footprints over the field of view of 43-km resolution on average (Kerr et al. 2010), while the three beams for Aquarius are approximately elliptical and have footprints of 76 × 94, 84 × 120, and 96 × 156 km2 (Lagerloef 2012)]. Therefore, if the ocean salinity field contains vertical gradients in the upper few meters, or if the ocean surface salinity has significant horizontal or temporal variability, there could be a physical difference between the satellite and in situ salinity values that would complicate calibration and validation of the satellite’s performance. The target defined for these satellite missions is to achieve a precision of 0.1–0.2 pss. This precision is sufficient to detect typical interannual SSS variability, such as that linked to El Niño–Southern Oscillation or to the Indian Ocean dipole, seasonal SSS variability in areas that have significant seasonal cycles [shown by Bingham et al. (2012) to cover 37% of the ocean surface between 60°N and 60°S and have a median seasonal SSS amplitude of 0.19 pss], mesoscale transport of salt by large eddies across strong fronts (Reul et al. 2014a; Kolodziejczyk et al. 2015), or intraseasonal SSS variability (Li et al. 2015, and references therein).
This paper synthesizes present knowledge of the processes that contribute to the formation and evolution of near-surface vertical salinity gradients and subfootprint-scale variability. The magnitude of these gradients is quantified whenever possible as a function of environmental conditions. The potential impact of both vertical salinity gradients and subfootprint-scale variability on satellite and in situ salinity data comparisons will be discussed.
VERTICAL STRATIFICATION AND SUBFOOTPRINT VARIATIONS.
Vertical stratification in the density of the upper ocean is controlled by the vertical profiles of temperature and salinity. Vertical stratification in temperature has been extensively studied over the past several decades, as it is responsible for observed differences in sea surface temperatures derived from infrared radiometers, microwave radiometers, and in situ measurements (Minnett and Kaiser-Weiss 2012, and references therein). In contrast, relatively few studies of upper-ocean salinity stratification [see the recent climatology discussed by Maes and O’Kane (2014)] have been performed. The addition of freshwater to the ocean surface (from precipitation, river runoff, or melting of sea ice) and the removal of freshwater (through evaporation) can generate vertical salinity gradients in the upper few meters of the ocean. Vertical stratification can be strong under low wind speed conditions when there is little mixing in the upper few meters of the ocean. When the wind speed at the ocean surface is greater than ∼6 m s–1, wind stress–induced momentum tends to homogenize the upper few meters of the ocean’s surface layer (Matthews et al. 2014). When cooling at the surface leads to unstable density stratification, as typically happens at nighttime, convective overturning can also generate a well-mixed surface layer. Regardless of the source of the mixing, salinity is homogeneous throughout the well-mixed layer; this homogeneous condition is considered to be the “normal” case, characterized by a salinity profile that is constant with depth, as shown in Fig. 2a. For the normal condition, radiometrically measured salinity is expected to be comparable to in situ salinity anywhere in the near-surface layer. The sections titled “Rain freshening,” “Freshwater plumes,” and “Evaporation” discuss the processes that lead to surface freshening (i.e., negative surface salinity anomalies; Fig. 2b) and surface salinification (i.e., positive surface salinity anomalies; Fig. 2c). Based on observational salinity data, the magnitudes of vertical salinity gradients during these conditions will be estimated. These processes, in particular rain freshening and freshwater plumes, are also associated with strong horizontal variability. In the section titled “Subfootprint variability,” we discuss subfootprint variability in a wider context.
Salinity in the upper ocean, especially in the ocean surface boundary layer (OSBL), is subject to large spatial and temporal variability due to various contributing processes, including freshwater influx from precipitation. The scales of this variability, though not well understood or quantified, are assumed to be related to the modification of the freshwater input by the air–sea fluxes of heat and momentum, upper-ocean mixing, and advection. Low-latitude ocean regions characterized by strong rainfall, low to moderate surface winds, and high advection are therefore expected to display relatively strong spatial and temporal variability in SSS.
Under normal conditions when no rainfall is present, the OSBL is characterized by a nearly uniform density, active vertical mixing, and a high rate of turbulence dissipation (Stevens et al. 2011; Sutherland et al. 2014a). As a result, vertical salinity gradients in the upper 10 m are expected to be small (Henocq et al. 2010; Anderson and Riser 2014). In regions where normal conditions dominate, it is appropriate to neglect vertical salinity gradients when using Argo for large-scale validation of SMOS and Aquarius SSS. However, in cases where rainfall induces a near-surface vertical salinity gradient, it is possible that salinity measurements at depths of a few meters might not accurately reflect SSS measured by the satellite in the upper few centimeters. Therefore, Argo measurements made at a few meters depth might not be suitable for validating satellite measurements of SSS.
When averaged globally, rain-induced salinity stratification of the upper mixed layer creates a bias of about –0.02 pss between the salinity measured at a few centimeters and at a few meters. Regional averaging shows that this bias increases to –0.03 pss in the tropics (Drucker and Riser 2014). Rain-induced salinity anomalies and near-surface haloclines resulting from individual rain events were extensively observed in the western Pacific warm pool during the Tropical Ocean and Global Atmosphere (TOGA) Coupled Ocean–Atmosphere Response Experiment (COARE; Soloviev and Lukas 1996, 1997) and in the Bay of Bengal during the Joint Air–Sea Monsoon Interaction Experiment (JASMINE; Webster et al. 2002). More recently, vertical salinity gradients between the upper centimeters and a few meters depth have been observed by Argo surface temperature salinity (STS) profilers (Anderson and Riser 2014), the Air–Sea Interaction Profiler (ASIP; Ward et al. 2014; Walesby et al. 2015a; Sutherland et al. 2014b), the towed Surface Salinity Profiler (SSP; Asher et al. 2014a), shipboard thermosalinographs (TSGs) at two depths (Asher et al. 2014a), and surface drifters (Reverdin et al. 2012).
While near-surface vertical salinity gradients from individual rain-induced freshening events can be large (>1 pss between a few centimeters and a few meters), the distribution of rain events in both space and time is relatively sparse, even in regions characterized by high rainfall. For example, several recent studies have estimated that on average rain-induced surface freshening occurs ∼12% of the time when considering the global ocean and ∼16% of the time when considering the tropics (Boutin et al. 2013; Anderson and Riser 2014; Drucker and Riser 2014; Meissner et al. 2014). Similarly, Anderson and Riser (2014), using Argo STS float measurements, found that salinity in the upper 4 m is, in most cases, well mixed (i.e., the difference between salinity at a few centimeters and 4 m is less than 0.1 pss for 97% of the observations). Observational studies have consistently shown that, in most cases, near-surface fresh anomalies produced by rainfall are eliminated quickly (typically within a few hours) by mixing, advection, and vertical convection. For example, the deepening of fresh cells to 40-m depth has been observed in the five hours after rainfall with a surface freshening signature of 0.12 pss (Wijesekera et al. 1999; Soloviev et al. 2002). On the other hand, Walesby et al. (2015a) observed a fresh lens that persisted for more than 15 hours, with little background mixing. The processes governing the vertical and horizontal evolution of fresh lenses are not well understood.
Several studies have attempted to quantify the difference between satellite and in situ salinity to determine the value of the rain freshening effect ΔS (pss) as a function of rainfall rate R (mm h–1) and time since rainfall. Unfortunately, both rain- and wind-generated roughness increase the microwave emissivity of the sea surface, mimicking a decrease in satellite-derived salinity measurements. Consequently, the effect of increased roughness must be addressed before determining the freshening due to rain. Although the effect of wind on roughness (and microwave emissivity) is relatively well known, rain-induced roughness is less understood. The Aquarius instrument is useful for studying this problem because the collocated L-band radiometer and L-band scatterometer are both sensitive to changes in surface roughness, whereas the scatterometer is insensitive to changes in salinity, thereby providing the means for isolating the effects due to surface roughness. Comparison of the signals from the two instruments suggests that the increase in emissivity due to rain-generated roughness is significant at low wind speeds (Tang et al. 2013). At moderate and higher wind speeds, however, rain-generated roughness does not appear to be a major component of the total roughness (Tang et al. 2013; Boutin et al. 2014; Meissner et al. 2014).
Once roughness and atmospheric effects are removed, comparing SSS measured by SMOS (Boutin et al. 2014) or Aquarius (Drucker and Riser 2014; Meissner et al. 2014; Santos-Garcia et al. 2014) to collocated in situ salinities not under the direct influence of instantaneous rainfall shows that ΔS/R induced by rainfall is estimated to be around –0.15 pss (mm h–1)−1 (Table 1). However, this bias is an average obtained under a range of environmental conditions (wind, rain history, stratification, net heat flux, etc.). Available direct measurements of ΔS under different conditions suggest that it is unlikely that DS for a particular rain event can be accurately predicted solely by R.
Figure 3 shows the dependence of ΔS on R for the range of ΔS/R (Table 1) found from satellite SSS. Also shown is ΔS as a function of R calculated by Schlüssel et al. (1997) as part of TOGA COARE. Schlüssel et al. (1997) determined this relationship for the salinity difference between the molecular diffusion sublayer (about 50 µm) and the bulk salinity, taking into account the effects of the near-surface mixing induced by raindrops. Despite the variability found in the experimental data, the trends derived from the model and the trends derived from satellite measurements agree well. This convergence of results and theory suggests the value of –0.15 pss (mm h–1)−1 is relatively robust. However, at present there are no concurrent collocated in situ measurements of R and near-surface salinity profiles that include the radiometric sampling depth that can be compared with satellite-derived estimates of ΔS.
Although the molecular skin layer modeled by Schlüssel et al. (1997) is much thinner than the radiometric measurement depth used to define ΔS, it is reasonable to equate salinity across the two depths and expect the model to provide an estimate of ΔS. First, Schlüssel et al. (1997) compared their model results to salinity measured between depths of 2 and 3 cm during several rain events. They found that the magnitude of the measured salinity decrease in the upper few centimeters was consistent with their model predictions for the molecular skin layer. Second, Schlüssel et al. (1997) hypothesized that the very near surface is rapidly homogenized when near-surface mixing caused by the impact of raindrops is taken into account. This idea is consistent with the bubble population measurements made by Ho et al. (2000) that show that during rain events the upper few centimeters of the water surface are well mixed.
Further work is needed to resolve the minimum in situ sampling depth that is required to fully resolve the near-surface salinity profile. Ideally, the profile would sample up to the radiometric penetration depth (i.e., 0.01 m). As noted above, however, the kinetic energy imparted to the water surface by raindrops homogenizes the top few centimeters (Ho et al. 2000), implying that the surface is well mixed at least to the radiometric depth. Therefore, techniques that resolve the top few centimeters should be sufficient. Nevertheless, in highly resolved vertical salinity profiles measured with ASIP during two rain events, gradients larger than 0.1 pss between the sea surface at a few centimeters depth and 30-cm depth have been observed (Ward et al. 2014). ASIP profiles also show that rainfall quickly stratifies the OSBL, inhibiting turbulence. This stratification may lead to strong gradients after the rain has ceased.
Rain-induced surface freshening and the resulting stratification appear to depend nonlinearly on the freshwater input volume, the strength and direction of the surface heat fluxes, and wind-induced mixing. Asher et al. (2014a) developed a one-dimensional diffusion model that fit observed vertical salinity profiles for the top 2 m and direct measurements of R to modeled salinity profiles by tuning the turbulent diffusivity coefficient and a scale depth for mixing. This model provided the basis for developing a macroscale Rain Impact Model (RIM) by Santos-Garcia et al. (2014). RIM was developed using Aquarius data and Hybrid Coordinate Ocean Model (HYCOM) output (at ∼10-m depth) from the Pacific intertropical convergence zone. It estimates the impact on Aquarius SSS based on rain accumulation over the previous 24 h and time since rainfall. The authors show that the difference between 10-m salinity and salinity measured by Aquarius is not only sensitive to R when the satellite is overhead but also to the rain history over the past 25 h, especially when wind speed is low.
In addition to rain, other sources of freshwater to the surface ocean are river discharge and melting ice. Freshwater plumes from rivers can contribute to the formation and evolution of barrier layers (Sprintall and Tomczak 1992; Pailler et al. 1999; Mignot et al. 2007; Reul et al. 2014b). In situ measurements (e.g., hydrographic ship surveys, Argo floats, voluntary observations from buckets, or TSGs on commercial ships) are too sparse in both space and time to allow full characterization of the generation and evolution of freshwater plumes. Studies have attempted to use ocean color data from the Coastal Zone Color Scanner (CZCS) (Longhurst 1993; Muller-Karger et al. 1995) or the Sea-viewing Wide Field-of-view Sensor (SeaWiFS; Fratantoni and Glickson 2002) to monitor the Amazon River freshwater plume. Reul et al. (2009) demonstrated the first satellite SSS retrieval by using the Advanced Microwave Scanning Radiometer for Earth Observing System (AMSR-E) C-band and X-band channels at 6.9 and 10.7 GHz, respectively, to measure SSS in the Amazon plume. More recently, SMOS and Aquarius data have been used to detect and characterize freshwater plumes for the outflows of the Congo (Hopkins et al. 2013; Reul et al. 2014c; Chao et al. 2015), the Mississippi (Gierach et al. 2013), the La Plata (Guerrero et al. 2014), and the Amazon (Grodsky et al. 2012; Reul et al. 2014b; Fournier et al. 2015; Korosov et al. 2015).
Significant spatial and temporal variability of SSS associated with river plumes can be detected using satellite SSS in regions with large river outflows (Fig. 4). As discussed for rain, near-surface vertical salinity gradients created by freshwater plumes can complicate the comparison of satellite and in situ salinity measurements, as seen with the 2–5 pss m−1 difference across the halocline shown by Lentz and Limeburner (1995). Plumes can also cause horizontal salinity gradients with spatial scales smaller than the footprint of the radiometers. Typical horizontal SSS gradients for the plumes from the Amazon (Lentz and Limeburner 1995) or Congo (Chao et al. 2015) exceed 0.2 pss km–1 and extend more than 250 km from the river mouth. Therefore, in the vicinity of a river plume, a spatially sparse array of in situ sensors can indicate very different SSS variability than a satellite sensor. High-frequency SSS variations (e.g., tidal effects) can be undersampled by satellite-derived SSS products due to the relatively long revisit time of the satellite (2–3 days for SMOS and 7 days for Aquarius).
At high latitudes, freshwater plumes can be caused by melting sea ice or by meltwater runoff from ice sheets, as has been observed in the seas around Greenland (Straneo and Heimbach 2013). Some of the freshest ocean waters are found in narrow high-latitude coastal currents such as the East Greenland Current. Meltwater from the Antarctic Ice Sheet is the main source of freshwater plumes in the Southern Ocean (Nicholls et al. 2009). While the direct impact of runoff on the coastal currents may be difficult to observe with satellite instruments, model simulations have shown that large meltwater runoff from the Greenland Ice Sheet changes the salinity of the seas surrounding Greenland (Marsh et al. 2010). Although these regions are poorly sampled by in situ observations, SSS retrievals for these areas from satellite L-band radiometers are now routinely available (Brucker et al. 2014a). Unfortunately, the study of freshwater at high latitudes is hindered by the presence of sea ice and icebergs (Brucker et al. 2014b), low water temperature (which reduces the L-band salinity signal-to-noise ratio, thus degrading SSS retrievals), and the prevalence of very rough sea surfaces. Despite these challenges, satellites provide regular monitoring capabilities for SSS that are critically lacking with in situ measurements.
Evaporation can also create vertical salinity gradients in the surface layer and, thereby, potential differences between satellite and in situ salinity measurements (Saunders 1967; Katsaros and Buettner 1969; Soloviev and Lukas 1997; Henocq et al. 2010; Anderson and Riser 2014; Drucker and Riser 2014; Asher et al. 2014b). The nature of the evaporation process and its impact, however, differs from that of precipitation in two major ways. First, evaporation increases salinity and cools the surface waters, both of which serve to increase density. This weakens or destabilizes the density gradient, thereby potentially initiating convective mixing (Yu 2010; Asher et al. 2014b; Soloviev and Lukas 2014). Conversely, precipitation freshens surface waters, reducing density, thereby strengthening the surface stratification and sustaining the freshwater lenses formed by rain (Schlüssel et al. 1997; Henocq et al. 2010; Boutin et al. 2013). Second, evaporation is almost always present, whereas precipitation occurs mostly as episodic events, although the surface freshwater volume flux during rain episodes greatly exceeds the flux due to evaporation over a time period equal to the rain event. During the TOGA COARE field experiments, Lin and Johnson (1996) observed that precipitation rates are highly variable, with peaks of 20 mm day–1, while evaporation rates are more stable, consistently around 3–5 mm day–1. The small volume flux, together with the destabilizing effect on the vertical density profile, implies that evaporation-induced surface salt enrichment (positive salinity anomalies) is relatively weak and short-lived (Yu 2010).
The magnitude of evaporation-induced positive salinity anomalies depends on both evaporation intensity and surface turbulence. Two processes can produce salt increase under evaporation: the salinity skin effect due to near-surface diffusive processes and the daily diurnal cycle in sea surface temperature. Saunders (1967) derived a parameterization for the change of salinity ΔSskin across the salinity skin layer by scaling the layer thickness as the one-third power of the diffusivity. The mutual enhancement between evaporation and wind led him to conclude that the ΔSskin is at most 2%, or around 0.07 pss for a surface salinity of 35 pss, in the extreme condition of low wind speed and large difference in air–sea specific humidity. In support of this result, Fedorov et al. (1979) obtained an estimate of DSskin = 0.12 pss from a laboratory experiment. Yu (2010) produced a global estimate of ΔSskin and suggested a magnitude of 0.05–0.15 pss. Given that the salinity skin layer is typically less than 0.1 mm thick (Zhang and Zhang 2012) and in situ instruments typically measure salinity and temperature deeper than 2 mm below the sea surface (Soloviev and Lukas 2014; Reverdin et al. 2013; Anderson and Riser 2014; Fig. 1), the salinity variations in the skin layer cannot be observed at sea. Nevertheless, Yu (2010) suggested that the salt increment in the skin layer is not a major source of error, because the salty skin layer is usually accompanied by a cooling of 0.2°–0.5°C, which is statically unstable and subject to convective overturn.
Soloviev and Lukas (1997) suggested that continuous evaporation can cause salinity to increase in the diurnal mixed layer, because the positive buoyancy flux due to diurnal heating promotes stable stratification by suppressing turbulent mixing with the water below (see Fig. 5). Confirmation of this hypothesis is provided by several field studies that have documented the existence of a relatively small salt-enriched diurnal cycle that is present under light winds (Soloviev and Lukas 1997; Asher et al. 2014b; Drushka et al. 2014; Hodges and Fratantoni 2014). Asher et al. (2014b) reported that salt-enhanced diurnal surface lenses (0.01–0.05 pss) around 0.5 m thick are common in the subtropical North Atlantic when wind speeds are less than 4 m s–1 and the average daily insolation is greater than 300 W m–2. In most cases, however, the magnitude of the salinity increase is usually small, comparable to the uncertainty in the measurements (Soloviev and Lukas 2014; Anderson and Riser 2014). Thus, it is postulated that diurnal salinity anomalies are also unlikely to induce significant biases between radiometrically measured salinities and salinities measured at depths of a few meters (Asher et al. 2014b).
A satellite measurement of SSS represents a near-instantaneous spatial average of the surface salinity field weighted by a function related to the satellite antenna pattern over a characteristic scale that is given by the satellite footprint. If SSS is uniform over the spatial scales averaged by a satellite, then a single in situ salinity measurement anywhere within the satellite footprint provides an accurate ground truth measurement that is representative of the remotely sensed value. However, model simulations (Johannessen et al. 2002) have shown that the salinity field is in some places spatially or temporally inhomogeneous, so that the relationship between the instantaneous, spatially averaged salinity measured by satellite and a single in situ measurement within the satellite footprint is not well understood. In ocean regions characterized by horizontal variability with spatial scales less than the satellite footprint, the subfootprint variability could be a source of difference between satellite and in situ data. When comparing salinity data taken at a point to the spatially averaged value reported by satellite, the SSS variability within the satellite footprint (i.e., subfootprint variability) may need to be taken into consideration.
Commonly used data for purposes of calibrating and validating satellite salinity measurements are those provided by surface drifters, moored buoys in the Tropical Atmosphere Ocean/Triangle Trans-Ocean Buoy Network (TAO/TRITON), or the Prediction and Research Moored Array in the Tropical Atlantic (PIRATA) array, and, most notably, the Argo array (which as of 28 July 2015 contains 3,881 profiling floats and produces the only near-synoptic observations of upper-ocean salinity throughout the World Ocean). However, the approximate 3° × 3° spacing of Argo profilers is a factor of 3 larger than the scale of the satellite footprints, which means that Argo does not resolve subfootprint-scale horizontal variability. Similarly, neither TAO/TRITON nor PIRATA can resolve subfootprint-scale variability in SSS.
In contrast to vertical gradients, subfootprint-scale variability in SSS can exist at all wind speeds. Both mesoscale and submesoscale features in the ocean that are responsible for the subfootprint variability of SSS are driven in large part by internal variability associated with ocean circulation. In fact, there can be significant horizontal variability on these larger scales that is not associated with vertical stratification in the upper few meters of the ocean.
Existing observational and modeling studies have provided some understanding of mesoscale and submesoscale SSS variability. For example, in a study comparing in situ data and the output of a high-resolution Massachusetts Institute of Technology (MIT) model of the Atlantic Ocean, Sena Martins et al. (2015) have shown that the annual cycle of SSS explains up to 70% of the total variability observed in some regions of the tropical Atlantic. However, this implies that in most regions at least 30% of variability is on scales other than the seasonal cycle. In fact, Sena Martins et al. (2015) show that SSS variability on time scales shorter than 30 days exceeds 0.1 pss in 42% of the 1° × 1° grid boxes of the model. When the annual cycle is subtracted, the temporal scales of the short-term variability in the model are 4–5 days throughout the Atlantic Ocean (confirmed by results from several mooring stations), and the spatial scales vary between 10 and 150 km. Delcroix et al. (2005) used TSG measurements from the Voluntary Observing Ship (VOS) Program that have 1–3-km resolution, as well as TAO-TRITON and PIRATA moorings at daily resolution, to estimate small-scale SSS variability in the tropical oceans. They reported the mean SSS variability in 2° (longitude) × 1° (latitude) boxes over 10-day intervals to be approximately 0.2 pss. However, there are ocean regions that are characterized by much stronger spatial variability. For example, Maes et al. (2013) analyzed TSG data from the Coral Sea and reported SSS variability as large as 0.6–1 pss over spatial scales of 100 km.
Quality-controlled TSG data (Delcroix et al. 2010; Alory et al. 2015) provide a new, improved resource for estimating subfootprint, near-surface salinity variability (recognizing that TSGs typically sample at 3–7-m depth, depending on the ship, and so do not necessarily represent salinity measured by satellites). Figure 6 shows the analysis of salinity variability derived from the standard deviation within 100-km intervals along a TSG track s100km (data produced by Alory et al. 2015; www.legos.obs-mip.fr/observations/sss). The σ100km values were then binned into 2° × 2° grid boxes, and the 95th percentile value σ95 (i.e., the 95% level of the cumulative distribution of σ100km) was computed. Because the distribution of standard deviations within each grid box is not necessarily Gaussian, the average of σ100km in a grid box does not necessarily represent the typical variability. Therefore, σ95 is shown as it represents an upper bound on the variability. Figure 6a provides a map of σ95, and Fig. 6b shows a histogram of σ95, along with the cumulative distributions of σ95 and of 2 × σ100km (which for a Gaussian distribution of σ100km would contain 95% of the points) overlaid. The σ95 histogram (Fig. 6b) shows a median value of 0.12 pss for the ocean regions in Fig. 6a. The cumulative distribution of σ95 is shifted to slightly larger values compared to that of 2 × σ100km because s100km is skewed toward large values (Fig. 6b). However, both cumulative distributions show that in about 25% of the cases SSS spatial variability exceeds 0.15 pss over 100-km scales, and in about 10% of the cases it exceeds 0.25 pss.
Detailed analysis of σ95 regional differences (Fig. 6a) indicates that SSS spatial variability exceeds 0.5 pss in regions affected by western boundary currents, major river plumes (e.g., the Amazon), and several coastal regions, demonstrating that, in many regions, subfootprint-scale SSS variability is larger than 0.1 pss. It should be noted that the instantaneous SSS variability may differ from this map as it was made by combining variability observed during different seasons and years. Finally, patterns of subfootprint variability derived from TSG data agree with the analysis of a HYCOM ocean data assimilation product (which excluded TSG data), conducted by Vinogradova and Ponte (2012). Vinogradova and Ponte (2013) quantified SSS variability within 1° × 1° bins to be as high as 0.2 pss near western boundary currents and in river outflow regions.
EMERGING TECHNOLOGY TO MEASURE NEAR-SURFACE SALINITY.
L-band microwave radiometers measure salinity in the top few centimeters of the water column. Development of in situ platforms and instruments that are capable of measuring salinity at these shallow depths is a very active field of research.
On global scales, most near-surface salinity data are from the Argo profiler network. Argo floats measure salinity using a conductivity–temperature–depth (CTD) sensor with a typical uppermost measurement depth of between 3 and 5 m (Boutin and Martin 2006). This depth is set in order to avoid ingesting sea surface contaminants into the CTD sensor, since these contaminants would degrade sensor stability over the life span of the Argo float.
The STS sensor has recently been developed and implemented on some Argo floats (Anderson and Riser 2014; Riser et al. 2015). An STS-equipped Argo float contains a second, free-flushed, conductivity sensor that is used in conjunction with the standard CTD sensor. The STS sensor samples at 1 Hz concurrently with the standard CTD, both near the float parking depth (980–960 dbar) and again in the upper ocean (20–3 dbar) just before the standard CTD sensor is turned off. After the CTD sensor turns off, the STS sensor continues sampling as the float progresses through the ocean surface, continuing for approximately 500 s as the float prepares to transmit data. Because the STS sensor measures through the film of the ocean surface, its calibration is expected to drift due to fouling. To correct for drift, STS conductivity data are scaled to agree with the mean conductivity from the reference CTD for a region with a small temperature gradient. The resultant mean STS-derived salinity is within 0.01 pss of the reference salinity along the entire profile (Anderson and Riser 2014).
During the first Salinity Processes in the Upper Ocean Regional Study (SPURS-1) field experiment (Lindstrom et al. 2015), multiple platforms were deployed and tested, including a mooring with CTD sensors installed at depths of 0.86 and 2.1 m (Farrar et al. 2015); drifters measuring at depths of 0.5 (Centurioni et al. 2015) and 0.2 m (Reverdin et al. 2015); Wave Gliders with CTDs mounted at 0.3 and 8 m (Hodges and Fratantoni 2014); a “salinity snake” that measures salinity in the top few centimeters of the ocean (Schanze et al. 2014; Paulson and Lagerloef 1993); a surface-following towed profiler that measures salinity and temperature at four fixed depths in the upper 2 m of the ocean, with a minimum measurement depth of 0.1 m (Asher et al. 2014a,b); and ASIP, which provides vertical profiles of temperature and salinity in the upper 50 m of the ocean with vertical resolution on the order of a few centimeters and an upper depth of 0.02 m (Fig. 1). Results from the SPURS-1 field experiment are very useful to contrast these different platforms in their abilities to measure the near-surface stratification. For example, in situ platforms measuring at a single point (e.g., ASIP, STS–Argo, and Argo) undersample in terms of area coverage (except in the very special case where two adjacent Argo profilers surface at the same time when a satellite is overhead) and have time scales much longer than the satellite revisit times. Moving instruments (e.g., the Salinity Snake and ship-mounted TSGs) have better spatial coverage, but the data they provide may not be coincident or cotemporaneous with the satellite.
In situ measurement of salinity in the top few centimeters of the ocean is difficult: on nonwave-following platforms, ocean surface vertical motion due to waves advects the water in the desired sampling region past the sensor faster than the response time of most commonly used conductivity-based salinity probes. Even platforms designed to follow large-scale wave motions at the surface have integrated measurement depths of a few centimeters (Fig. 1). Furthermore, conductivity-based salinity estimates are sensitive to the presence of bubbles, and the probes are often sensitive to fouling by biofilms, both of which are prevalent close to the sea surface. Existing methods to measure the near-surface salinity will be improved and new technologies will be developed during future field experiments (e.g., SPURS-2).
SUMMARY AND RECOMMENDATIONS.
The spatiotemporal variability of SSS within a satellite footprint (50–150 km) is a major issue for satellite SSS validation in the vicinity of river plumes, frontal zones, and significant precipitation. In other regions, while much reduced, this variability is often nonnegligible: in 65% of the grid boxes regularly observed by ships of opportunity (Delcroix et al. 2010), the SSS standard deviation along a 100-km transect reaches 0.1 pss. Hence, in many satellite–in situ comparisons, it is of primary importance to account for SSS variability within a satellite footprint. Information on the probability distribution function of SSS in satellite footprints is required, as are autocorrelation statistics such as those determined in some regions by Delcroix et al. (2005). Unfortunately, this variability remains very poorly documented due to the vast undersampling of the majority of the World Ocean (Fig. 6a). Clearly, knowledge of mesoscale and submesoscale SSS variability needs to be improved in terms of magnitude, spatiotemporal distribution, and related dynamics and impacts. In particular, high-resolution in situ measurements must be made in regions of strong variability. Future field campaigns such as SPURS-2 in the eastern tropical Pacific low-salinity region will enhance our understanding of small-scale SSS variability and related dynamical processes in rain-dominated regions.
Although NASA’s Soil Moisture Active Passive (SMAP) mission has a primary objective to measure soil moisture, it is possible to use SMAP data to retrieve salinity and improve the spatial sampling of SSS. The upcoming Surface Water and Ocean Topography (SWOT) satellite, to be launched in 2020, will provide sea level (and therefore derived geostrophic current) measurements that will resolve features with a wavelength of 15–100 km, which may facilitate the study of SSS variability on small scales. Emerging high-resolution modeling efforts will also give new insight into the dynamics of mesoscale and submesoscale variability of SSS. Although horizontal salinity variations are more likely to affect comparisons of satellite and in situ salinity, rainfall can in some cases produce vertical salinity gradients exceeding 1 pss m–1; consequently, it is recommended that satellite and in situ SSS measurements less than 3–6 h after rain events should be considered with care when used in satellite calibration/validation analyses. Satellite SSS measurements can be expected to improve in the future, so a detailed understanding of the processes that generate and control the evolution and fate of rain-induced surface freshening events is necessary in order to optimize the use of both satellite and in situ salinity observations. Future studies should, therefore, concentrate on characterizing the vertical salinity profile between the ocean surface and 10-m depth, the penetration of raindrops within the ocean, the effects of splashing and mixing by raindrops, and the small-scale horizontal and vertical advection of freshwater anomalies at the ocean surface. Some Argo profilers enable sampling the upper 3 m of the ocean. Such efforts should be encouraged, including efforts to assess the quality of these new, near-surface measurements. Furthermore, because these processes are coupled to the air–sea fluxes of heat and momentum, it would be advantageous and prudent to perform these assessments under a range of forcing conditions, with particular attention to characterizing necessary ancillary information such as the droplet size spectrum and surface heat fluxes. Ideally, models of salinity stratification in response to precipitation, wind, and advection should reconcile surface and near-surface observations. Parameterizing the near-surface salinity stratification into a global ocean circulation model has been attempted and has shown encouraging results in comparisons with Aquarius SSS and Argo 5–10-m salinity (Moon and Song 2014; Song et al. 2015). Understanding these phenomena at the scales of both an individual rain event and a satellite pixel will help improve the parameterization of rainfall in computational fluid dynamics models.
For the upcoming SPURS-2 field experiment in 2016/17, and looking into the future, new robust salinity sensors are required. For profiling platforms, high spatial resolution is needed. For fixed-depth platforms, better quantification is needed of the actual depth range sampled by the sensor, as well as minimizing platform issues such as flow perturbations and vertical averaging.
This paper was developed from discussions of the Satellite and In Situ Salinity (SISS) mailing list (http://siss.locean-ipsl.upmc.fr) and several SISS-sponsored working group meetings at the American Geophysical Union (AGU) Fall Meeting and European Geosciences Union (EGU) General Assembly. We thank the following people who read the manuscript and provided comments: Eric Bayler, Rafael J. Catany, Meike Sena Martins, and Yuhe Tony Song.
CURRENT AFFILIATION: Laboratoire de Physique des Océan (CNRS-IRD-UBO-Ifremer), Plouzané, France