Seasonal Phase Relationships between Sea Surface Salinity, Surface Freshwater Forcing, and Ocean Surface Processes

Abstract

Sea surface salinity (SSS) can change as a result of surface freshwater forcing (FWF) or internal ocean processes such as upwelling or advection. SSS should follow FWF by ¼ cycle, or 3 months, if FWF is the primary process controlling it at the seasonal scale. In this paper, we compare the phase relationship between SSS and FWF (i.e., evaporation minus precipitation over mixed layer depth) over the global (non-Arctic) ocean using in situ SSS and satellite evaporation and precipitation. We found that, instead of the expected 3-month delay between SSS and FWF, the delay is mostly closer to 1–2 months, with SSS peaking too soon relative to FWF. We then computed monthly vertical entrainment and horizontal advection terms of the upper ocean salinity balance equation and added their contributions to the phase of the FWF. The addition of these processes to the seasonal upper ocean salinity balance leads to the phase difference between SSS and the forcing processes being closer to the expected value. We conducted a similar computation with the amplitude of the seasonal SSS and the forcing terms, with less definitive results. The results of this study highlight the important role that ocean processes play in the global freshwater cycle at the seasonal scale.

1. Introduction

The lower atmosphere and upper ocean are connected through fluxes of freshwater that freshen or salt the ocean or add or subtract moisture from the atmosphere through evaporation and precipitation. These fluxes occur on many different time and space scales, and can help drive the circulation of both the atmosphere and ocean through the transfer of buoyancy and latent heat [1]. One of the most important time scales for ocean surface freshwater fluxes is the seasonal one [2], which is related to such important phenomena as monsoon circulation and atmospheric rivers [3]. The atmosphere lifts water off the surface of the ocean on a seasonal basis and carries it either onto land or to a different part of the ocean. The globally averaged amplitude of this extraction of water is something like 10 mm of equivalent sea surface height as measured by ocean mass [4,5], altimetry [6], or ocean salinity [7]. It is likely, however, that the amplitude of the seasonal extraction is much larger in some regions than others, as the atmosphere transports water on a seasonal basis from one part of the ocean, say one hemisphere or ocean basin, to another.

Because the atmosphere and ocean are so closely connected through freshwater fluxes, we expect those fluxes to be reflected in the properties of the surface ocean, namely the salinity. As the ocean surface freshwater flux (evaporation minus precipitation, E-P) increases or decreases throughout the annual cycle, the surface salinity should increase or decrease as well with about a ¼ cycle, or three-month, phase delay [8]. This assumes that other ocean processes are not important on a seasonal time scale. However, other processes are important for determining surface salinity [9], particularly Ekman and geostrophic transport, and vertical mixing/entrainment. So, to the extent that the timing and amplitude of surface salinity changes do not match that of surface freshwater flux, the surface salinity may be determined by seasonal-scale processes internal to the ocean.

The seasonal cycle of the surface salinity of the ocean (SSS) has been the subject of a number of studies. At a global scale, patterns of amplitude and phase of the seasonal cycle of SSS have been presented [7,10,11,12,13], showing regions where the seasonal cycle is strong, i.e., the tropics, northern and southern Indian Ocean, Amazon and Congo River plumes, the Arctic, etc. SSS tends to be highest in the late winter/early spring months of March and September [7]. Maps of amplitude and phase have also been presented at a regional level for many locations, including the global tropics [14], global subtropics [15], Pacific basin [16], subtropical North Atlantic [17], Atlantic basin [18,19], tropical Atlantic [20], Indian basin [21], North Indian [22], and tropical Indian [23,24]. The result of these disparate studies is that we have a good idea of where the seasonal cycle of SSS is large, and how much variance it represents, over much of the globe. However, for the most part, these studies do not relate the amplitude and phase of the seasonal cycle to that of the forcing, specifically E-P.

Many other studies present the climatological seasonal balance of SSS as budgets (e.g., [8,9]), indicating the importance of the salinity tendency, gain and loss of freshwater at the surface, vertical and horizontal mixing, horizontal advection, etc. Again, these are usually conducted regionally, in such areas as the global subtropics [25], tropical Atlantic [26,27,28,29,30], subtropical North Atlantic [29,31,32,33], tropical Indian [21,34], south Indian subtropics [35], Southern Ocean [36], tropical Pacific [37,38], and South Pacific subtropics [39]. There are also a couple of studies of the salinity balance of the ocean globally [9,40]. In some of these studies, the seasonal budget is more or less closed (e.g., [31,34]), and in others it is not (e.g., [35]). When the SSS budget is climatologically closed, we would expect that the salinity tendency would peak at exactly the same time as the sum of all of the other terms, or we would expect the SSS itself to peak 3 months, or ¼ cycle, after the sum of the other terms [8,16,41]. However, this may not be the case everywhere, as [7] found that SSS systematically peaks about 1–2 months ahead of this expected three-month phase delay.

In this paper, what we are interested in is the amplitude and phase relationship between SSS and surface freshwater forcing on a seasonal basis, with additional modulation by internal ocean processes such as advection and vertical entrainment. Many of the studies just cited do address this question at a local level, but not on a global basis in a way we could use to quantify the large-scale movement of water off the ocean surface and onto land or to a different part of the ocean. As a typical example, Ref. [29] examined the SSS balance in three areas in the tropical and subtropical North Atlantic. They compared the salinity tendency with the sum of vertical advection, horizontal advection, and surface moisture flux in these areas. The results are mixed. In one area, in the central subtropical ocean, there is little seasonal variability of any of these quantities. In the other two areas, the western and tropical ocean, there are robust seasonal cycles, with the amplitude of the tendency being similar to those of the sum of the other terms. Most importantly for this work, there is a phase offset. The tendency peaks about 2 months earlier than the sum of the other terms. Phase offsets like this are common in these kinds of SSS seasonal budget evaluation studies (e.g., [36,39]). Often, these studies point to unexamined processes, such as entrainment to close the budget. Thus, it is important to evaluate these terms in the seasonal context.

As mentioned above, the phase and amplitude of the annual cycle of SSS or its tendency have been examined in many studies, but there has been little attempt to systematically relate these to that of the forcing or other terms in the upper ocean salinity balance. This was tried by [16], who found, as mentioned above, a systematic offset of about 1–2 months between peak SSS and peak surface forcing, versus the expected 3 months. They also found that the amplitude of the SSS seasonal cycle was larger than that of the surface forcing. This study was conducted with a crude, pre-satellite SSS field and mostly pre-Argo dataset. The SSS seasonal cycle has become much better understood over the past decade with the advent of the satellite measurement of SSS [10,42]. This and improved precipitation [43] and evaporation [44] datasets make it timely to examine the seasonal phase and amplitude relationships between SSS, surface freshwater forcing (FWF), and internal ocean processes. Of these two, we find the phase relationship more interesting and understandable from the datasets we have.

2. Data and Methods

The purpose of this study is to determine the phase and amplitude relationship between SSS and other terms of the salinity balance equation, i.e., surface freshwater forcing, advection, and vertical entrainment. Thus, we need to detail the data sources, method for computing the terms, and the method for computing the seasonal harmonics. The upper ocean salinity balance equation can be expressed in many forms. We followed the simple formulation of [8,16], i.e.,

$${\displaystyle \frac{\partial S}{\partial t}}={\displaystyle \frac{S_0(E-P)}{h}}-\overrightarrow{u}\cdot \nabla S-w{\displaystyle \frac{\partial S}{\partial z}},$$

(1)

where S is the upper ocean salinity, t is time, S₀ is a reference value of S set to 35, E is evaporation, P is precipitation, h is the mixed-layer depth, $\overrightarrow{u}$ is the horizontal velocity, w is the vertical velocity at the base of the mixed layer, and z is the vertical coordinate. The first term on the right-hand-side of (1) is the FWF, the second term is advection, and the third is the vertical entrainment or upward salt advection. We estimated these terms on a global basis using in situ and satellite data and determined their seasonal harmonics.

2.1. Data Sources

The datasets we used were:

• A gridded in situ salinity product derived from Argo floats (the “RG” data; [45,46]). This is a monthly product on a 1°X1° grid produced according to reference [45].
• OAFlux evaporation [44].
• MIMOC mixed-layer depth (Monthly Isopycnal and Mixed-layer Ocean Climatology; [47]).
• GPM IMERG precipitation [43].
• OSCAR (Ocean Surface Current Analyses Real-time) surface velocity [48].
• Ekman upwelling produced by the NOAA Coastwatch program. These are monthly 0.25° values of vertical velocity at the base of the mixed layer derived from wind stress data.

We averaged all data to a monthly 1° grid, with latitude ranging from 63° S to 64° N. The salinity and evaporation data were initially available every degree centered on half degree increments, and we regridded it to be centered on whole degree increments. The mixed-layer depth data were available every half degree, and we subsampled the points centered on whole degrees to obtain a consistent grid.

The MIMOC mixed-layer product we used is a climatology, not a time series. We tried some other time series mixed-layer products but found them not useful for our purposes. The MIMOC is derived from the same underlying source as the RG data. It is computed from the MLD taken from individual profiles and objectively mapped [47]. It is not the MLD derived from averaged profiles, which we believe would be less useful.

2.2. Computing Seasonal Harmonics

For the SSS, the FWF, and the FWF combined with advection and entrainment (the methods for computing these are detailed below), we calculated the harmonics in the same way, following [10,16]. We looped through all latitude and longitude values on the worldwide 1° grid and pulled out the time series at each location. We computed a regression on the annual and semiannual harmonics combined for each time series. The regression returned 5 coefficients, which we used to calculate the amplitude and phase of the annual and semiannual harmonics. We also calculated the r² value for both the annual and semiannual harmonic at each point, using the variance of the time series and of the annual and semiannual fits. We then calculated the f-statistic from the r² values, and the cumulative f-distribution from that. We considered fits with cumulative f-distribution values greater than 0.9 and with more than a year of total data points at a location significant. The vast majority of semiannual harmonic fits were not significant, and we focused on annual harmonic results in this paper. The results of this computation are the annual phase and amplitude of the fit at each location, as well as whether the fit is significant. An example is shown in Figure 1, with original SSS, entrainment/advection and FWF data, and the harmonic fits.

Suppose that the salinity is a seasonally varying sine function and the tendency is completely balanced by a seasonally varying FWF. Then, S is expressed as

$$S=Asin(\omega t+\phi ),$$

(2)

where ω is 2π radians/year, φ is the phase, t the time, and A the amplitude. The FWF is

$$FWF=Bsin(\omega t+\varnothing ),$$

(3)

where B is the amplitude of the FWF and ∅ is a different phase. Taking the derivative of S with respect to time, and setting ∂S/∂t and FWF equal to each other, we obtained

$$A\omega \cos \left(\omega t+\phi \right)=Bsin\left(\omega t+\varnothing \right).$$

(4)

For this to be balanced, we need ∅ = φ + π/2. That is, the FWF leads S by ¼ cycle. We also have

$$A={\displaystyle \frac{B}{\omega }}.$$

(5)

So, if we plot A, the amplitude of S vs. B, the amplitude of FWF for different regions of the ocean, the slope should be 1/ω, i.e., 1/2π.

2.3. Computing Advection and Entrainment

We calculated horizontal advection using the OSCAR dataset and the surface salinity. Horizontal advection is defined as the dot product of the surface velocity vector and the surface salinity gradient (Equation (1)). The OSCAR product consists of zonal and meridional components. Its time resolution is every 5 days; so, we calculated monthly averages by directly averaging all the components separately at a given location in a given month. The grid for ocean velocity was at a 1° resolution and centered on every half degree. To calculate the dot product, we needed the meridional and zonal partial derivatives of the surface salinity. The zonal partial derivative was calculated as the difference between the salinity at each grid point minus the salinity at the grid point with the previous longitude value, divided by the distance between them. This changed the centering in just the longitudinal direction to every half degree; thus, we regridded in the latitudinal direction so every point was centered on the half degree in both directions, to be consistent with the velocity data. We calculated the meridional derivative in the same manner, finding the difference between points with different latitudes along longitude lines, dividing by distance, and regridding to every half degree. We then calculated advection at every half degree by multiplying the zonal and meridional components of the salinity gradient and the surface velocity, and finally regridded the advection to be centered on every whole degree.

The mean advection (Figure 2b) indicates the transformation of surface water as it flows from one part of the ocean to another. It is generally small compared to the surface forcing (Figure 2a) and vertical entrainment (Figure 2c), though it can be larger locally. It is large within intense western boundary currents, such as the Gulf Stream, Kuroshio and Brazil–Falklands Confluence, and also in the Antarctic Circumpolar Current at 35–45° S. It results in both the freshening and salting of the surface layer. There are bands of alternating impacts in the tropics.

We defined vertical entrainment as the monthly vertical velocity multiplied by the annual average vertical salinity gradient added to the seasonal vertical salinity gradient multiplied by the annual average vertical velocity:

$$w{\displaystyle \frac{\partial S}{\partial z}}=w^{\prime }\overline{{\displaystyle \frac{\partial S}{\partial z}}}+\overline{w}{\displaystyle \frac{\partial S^{\prime }}{\partial z}}.$$

(6)

We ignored the term where the seasonal vertical velocity multiplies the seasonal vertical salinity gradient. This is the product of an annual-period sine function and a cosine, which, by trigonometric identity, makes a semi-annual period sine. We did not study semi-annual period variability in this paper.

To calculate the vertical velocity, we combined the upwelling and the vertical motion of the mixed layer base,

$$w^{\prime }=w_E+{\displaystyle \frac{\partial h}{\partial t}},$$

(7)

where w_E is the Ekman upwelling velocity.

We subsampled the upwelling velocity grid to obtain upwelling values at whole degrees in the latitude range we were interested in. There was one month of data, August 2013, that had obviously incorrect values; so, we set that those values to null. To calculate the vertical motion of the mixed-layer base at each location, we subtracted the previous month’s value from the current month’s value and divided by the length of time in between the two time indices in seconds. This changed the centering in time, and so, we then averaged every two monthly values again to revert to centering in the middle of months. We then combined the upwelling and vertical motion of the mixed-layer base to get the vertical velocity. Because only water moving up into the mixed layer from below changes the salinity of the water in the mixed layer [9,49], we only included positive vertical velocity in our calculations. We set all negative (downward) vertical velocity values to zero. We took the mean at each location across all the years analyzed for the average vertical velocity. We defined entrainment as a change in salinity, not a movement of water.

For the vertical salinity gradient, we compared the salinity at the surface and 30 m below the average mixed layer depth at each location from the RG data [16]. For the average gradient at each grid point, we took the mean of the salinity at the surface minus the mean below the mixed-layer depth, divided by the distance of 30 m below the mixed-layer depth. For the seasonal vertical salinity gradient, we also subtracted the salinity 30 m below the mixed-layer depth from the surface salinity and divided it by the depth, with the salinity still varying in time.

$${\displaystyle \frac{\partial S^{\prime }}{\partial z}}={\displaystyle \frac{S\left(0\right)-S(MLD+30)}{MLD+30}}.$$

(8)

The average entrainment (Figure 2c) shows that this process largely acts to increase the salinity of the mixed layer. It is especially large in the tropics, on the eastern and western boundary of the tropical Pacific and in the Indian Ocean. There are some areas, specifically in the mid-latitude ocean subtropics where entrainment acts to freshen the mixed layer, where it is underlain by fresher water. The magnitude of this is small compared to the intense upwelling of salty water that occurs in the tropics [50].

One term that we have neglected is the lateral induction, where properties can be added to or removed from the mixed layer by horizontal advection through a sloping base [51,52]. The annual mean of this term was computed by [53] for the South Pacific. It shows that the term is large for the equatorward side of the South Pacific Tropical Water near 20° S and sporadically large for the equatorward side of the Antarctic Circumpolar Current. Otherwise, the term is small or zero throughout the South Pacific basin relative to other subduction terms. No indication is provided by [53] as to how large the seasonal variability of the lateral induction might be in the South Pacific.

2.4. Computing Forcing

As stated above, the mixed-layer depth values were from a monthly average climatology. The picture of mean FWF (Figure 2a) can be compared to that of [54], who showed a very similar map of E-P. Large areas of excess precipitation are seen in the intertropical convergence zone in the Pacific and Atlantic, the South Pacific Convergence zone in the western South Pacific, the tropical Indian Ocean, and at high latitudes. Excess evaporation occurs over large areas of the mid-latitude subtropics.

2.5. Computation of Histograms

We computed the area of every one-degree box within the latitude and longitude ranges we are using to normalize the histograms by area. Each histogram shows the normalized areas of the ocean where the phase peaks in various months. We only plotted areas with statistically significant fits, and both the significance and the area were considered separated for the salinity, the forcing, and the forcing combined with advection and entrainment.

3. Results

The four scatter plots in Figure 3 show the annual phase of the surface salinity compared to the annual phase of various combinations of forcing terms. Figure 3a(b,c,d) has 18,402 (10,119, 17,387, 9351) points. Adding both advection and entrainment causes fewer locations within the ocean to have statistically significant annual harmonic fits.

Comparing just SSS and the FWF (Figure 3a), the majority of the global ocean does not have the FWF peaking with the expected 3-month phase lead over SSS. That is, one would expect the cloud of dots in Figure 3a to be centered around the black lines, which indicate 3 months of phase lag. Instead, the SSS systematically peaks too early in most areas by 1–2 months. This agrees with the result of [7]. Another interesting aspect to the distribution shown is that, for most of the ocean, SSS peaks in the spring in either hemisphere and the forcing peaks in the winter. This can also be seen in the maps of [7,10].

Adding other terms of the salinity balance equation can change the picture. Considering FWF minus advection (Figure 3b), the cloud of points in the figure does not move much. Vertical entrainment has more impact (Figure 3c). Including entrainment both by itself and along with advection does move the clusters of points closer to being centered around the 3-month expected lag. This suggests that entrainment does more than advection to delay the phase of the ocean’s response to FWF. Grid points in the high latitude southern hemisphere largely have a SSS peak in austral spring, a forcing peak in late winter, and a forcing combined with advection and entrainment peak in early to mid-winter. Grid points in the tropics have a salinity peak in boreal spring and a forcing peak in late winter, which does not change as much as some other areas of the ocean when considering advection and entrainment. This could be related to the fact that there are fewer significant harmonic fits in the tropics overall. In the high latitude northern hemisphere, salinity peaks in early boreal spring, forcing peaks in late winter, and forcing combined with advection and entrainment peaks in early winter.

The scatter plots in Figure 4 are formatted in the same manner as those of Figure 3, but with the annual harmonic amplitude plotted for surface salinity and various combinations of forcing terms. The black line shows the expected slope of 1/2π (Equation (5)). The patterns in amplitude are much less clear than the patterns in phase. While none of the scatter plots show the straightforward correspondence predicted by the salinity balance equation, it seems that, when not including advection (Figure 4a,c), there are more areas of the ocean that have a greater seasonal variation in SSS than would be expected given the freshwater forcing and entrainment. When advection is included, there are more areas of the ocean that have a smaller than expected seasonal salinity amplitude given the amplitude of the forcing terms including advection. In addition, adding advection causes the cloud of points to visually spread less out from the expected line in the scatter plots above. We tested this observation by computing the RMS deviation of the points in Figure 4 from the black line showing the expected values. These RMS differences were (0.130, 0.107, 0.112, and 0.107) for panels (a, b, c, and d), respectively. This shows significantly smaller differences when including the advection and entrainment. The relationship of amplitude to horizonal advection requires additional research and this paper focused largely on phase, but it is noteworthy that vertical entrainment seems to affect phase more and horizonal advection seems to affect amplitude more. Unlike with the phase, clear patterns with amplitude do not emerge as a function of the latitude.

The maps in Figure 5 show in more detail where in the ocean there are statistically significant seasonal harmonics and where the difference between the phase of sea surface salinity and various forcing terms is close to the expected 3 months. In all of these plots, there are relatively few significant fits around the equator and between 20° S and 40° S. The expected 3-month difference is in green. For the comparison between SSS and FWF only, large deviations from that occur in the eastern tropical basins, especially the Pacific, the Arabian Sea, a band around 20° N in the subtropical North Atlantic, and a few other scattered locations (Figure 5a). The bands around 20° N in the North Pacific and North Atlantic are the result of strong poleward Ekman transport of seasonally forced variability in the intertropical convergence zone [17,54]. When we include advection (Figure 5b), the number of locations with significant fits decreases sharply, mainly in the eastern tropical Pacific and Southern Ocean. The eastern tropical Pacific in particular is a region of intense upwelling (Figure 2c). Perhaps here, there is a balance between seasonal surface forcing and non-seasonal upwelling. Adding entrainment (Figure 5c) leads to many areas being closer to the 3-month expected difference, especially in the subtropical South Pacific and western and northern North Pacific. Finally, the inclusion of all terms (Figure 5d) again reduces the number of areas with significant fits, but shows that more areas have phase differences close to 3 months.

Figure 6 shows the amplitude ratio of various forcing terms to salinity where there are significant annual harmonic fits. The expected ratio of 2π is orange in the color scale of the maps. Unlike the maps of phase difference (Figure 5), the amplitude ratios are highly variable. In many areas, the amplitude of the forcing terms is large compared to SSS (e.g., the northwestern North Pacific, or the South Pacific Convergence Zone in the western South Pacific). In other areas, the SSS has a large amplitude compared to the forcing (e.g., the tropical Atlantic). Adding advection and entrainment does not lead to amplitude ratios being closer to the expected value (Figure 6b–d).

The area where SSS and various forcing terms peak is shown in Figure 7. As seen in the scatter plots of Figure 3, SSS peaks in mainly in March and September. This is true for all latitude ranges. For high southern latitudes (dark blue bars), there is a clear peak in September—spring. In the tropics (30° S–30° N, red bars), the peaks are in both March and September, but with less seasonal variation. For higher northern latitudes (yellow bars), the peak is in February and March, i.e., late winter/early spring.

With just FWF and no ocean dynamics (Figure 6a, lower bars), the FWF peaks in January and July. The July peak is associated with high southern latitudes (purple bars). In the tropics (green bars), the peaks are in January and July, and in higher northern latitudes, the peaks are in January. The higher latitude peaks indicate maxima in E-P, possibly due to the intense evaporation that occurs in western boundary current regions, such as the Gulf Stream or Kuroshio in winter [52,53].

As ocean dynamics are added (Figure 7b–d), the peaks in forcing shift to earlier months. The July peak in panel a especially shifts back in time to June when entrainment is included (panels c and d) but not as much for only advection (panel b). The January peak in panel a shifts back to December/January with the inclusion of entrainment and (to a lesser extent) advection. These shifts are visible at all latitude ranges.

4. Discussion

We computed the annual harmonics of various terms of the upper ocean salinity balance equation, focusing mainly on the phase relationship between the terms and the salinity itself. We briefly discussed the relationship of the amplitudes (Figure 4 and Figure 6), but found much less definitive results and leave the more in-depth study of that for future work.

For much of the ocean, where there is a significant annual cycle of SSS, it peaks in the spring at most latitudes (Figure 3 and Figure 7; [7]). As a whole, this timing of the peak of SSS is likely related to the timing of the peak of E-P, which mainly occurs in the winter at most latitudes (Figure 7a lower bars). Generally, this freshening and salting of the surface ocean associated with evaporation and precipitation define the major part of the global water cycle [55,56,57]. This cycle extracts water from the ocean surface and moves it via the atmosphere from one part of the ocean to another, or from ocean to land, only to have it return to the surface ocean via rainfall or river runoff. The mean global transport of freshwater has been observed and reported in [58,59], but not so with the seasonal cycle. The magnitude of this seasonal movement of water and its mechanisms are important subjects for future study.

The freshwater forcing has a clear annual cycle over many parts of the ocean, but is, in many places, out of balance with the SSS tendency (Figure 5a). When we added advection and entrainment, the other terms on the right-hand side of Equation (1), the phase is rectified and becomes closer to matching with that of the SSS. However, we also found that areas with a large seasonal variability were not statistically significant. This is mostly true for regions that are out of balance. That is, many of the areas that are pink and brown in Figure 5a disappear in Figure 5b–d, especially due to advection (Figure 5b). Advection, therefore, has the effect of moving water out of regions that are strongly seasonal in terms of forcing and smearing the annual cycle out.

The main conclusion of this paper is that ocean dynamics are required to make the seasonal balance of terms in the upper ocean salinity/freshwater budget close in terms of timing. In particular, vertical entrainment plays a major role in rectifying the seasonal cycle of SSS relative to FWF. As the mixed-layer base deepens in fall and winter as part of its annual cycle, it mixes water from the interior into the surface. In the tropics, much of the underlying water is saltier than that at the surface due to the presence of subtropical underwater [60,61]; so, this process increases the salinity of the mixed layer. As such, the results presented here highlight the importance of the subtropical cell of [50] in the global water cycle. Unfortunately, there are no observations of the seasonal variability of the subtropical cell, and there is little understanding of the role that this cell might play in the seasonal motion of freshwater.