How WebGCM works
Part 8: A moving sea

Wind, salt and ice: the heat the sea holds and carries, and how its water moves.

Part 7's wet planet lay on a sea without currents, a slab of water 10 m deep that warmed and cooled where it lay. The climate model's sea is deeper, it holds salt, and it moves. It stores the summer's heat and gives it back in winter, its water floats or sinks by its temperature and its salt, the wind drives it and the turning planet steers it, and in the polar seas it freezes.

The model builds its ocean as it builds its atmosphere, as a stack of layers on the grid of Part 3 moved by the rules of Part 4: 58 layers, held at fixed depths near the surface and following surfaces of equal density below. The air hands it heat, fresh water and the push of the wind, and it hands back the temperature and the thickness of its top layer, 3 m, the share of the sunlight that layer lets through to the water beneath, and under ice the heat it gives the ice's base, which is all the air ever sees of the sea.

This part follows the heat that one column of the model's sea stores through a year, the density of salty water and the layers the model sorts it into, the drift that the wind drives, the slope of the sea's surface, the gyres that walls make and the ice that caps the polar sea. The two sections after it set the whole sea moving over Part 7's planet and follow the heat it carries toward the poles. The first six figures run the model's own ocean or ice code live in your browser, and the few pieces written only for a figure are labeled where they appear; the moving sea of the last two sections was computed in advance.

The sea holds heat

The sea's first effect on the climate is the heat it holds. A cubic meter of seawater takes 4.08 MJ to warm by one kelvin, its volumetric heat capacity, about 3,400 times as much as a cubic meter of air near the ground. The whole column of air above a square meter, about 10,000 kg of it, holds as much heat for each kelvin as the top 2.5 m of the sea. Part 7's slab, 10 m deep, held the heat of four such columns.

The model's sea is far deeper, and its water lies in 58 layers. The top twelve are 3 m thick at the surface and grow to 7 m, and together reach 57 m. The 32 below them keep at least 8 to 10 m each, to 374 m, and the last 14 at least 40 to 500 m, to 3,019 m. Where the water allows, a layer is thicker than that minimum and follows a surface of equal density, which the next section turns to. None of the layers is a mixed layer. The sea's counterpart of the air's mixed layer in Part 1 is a depth the model measures after it has mixed the water, down to which the wind and the cooling of the surface have kept one temperature and one salinity. The top layer alone holds the heat of 1.2 air columns for each kelvin, and a mixed layer 41 m deep that of 16.

The figure runs the model's own ocean, all 58 layers, in one column through a year from the March equinox, begun as the model's own starting sea lies at 45°N: 13.0 °C and 34.74 psu at the surface, cooling with depth toward 2 °C and 34.7 psu over a scale of 400 m, a smooth profile whose mixed layer is 7 m deep. Its heating is written for this figure, not computed by the climate model: ±133 W/m² that follows the sunlight a sea at 45°N absorbs through the year, strongest at the June solstice, less 10 W/m² for each kelvin the surface is warmer than at the start. All of it counts as sunlight, of which the ocean lets 31 percent pass its top layer to warm the water beneath. A wind stress of 0.1 N/m² stirs the sea, its currents are held at rest, and Part 7's slab takes the same heating.

The start is a smooth profile, not a mixed layer, so the wind first mixes it: by day 10 the water is uniform to 53 m and the surface has cooled from 13.0 to 12.3 °C. Then the heating outruns the wind. Warmed water at the surface is lighter than the water beneath it, and the wind's work can stir it only so deep: from day 20 the mixed layer shoals from 55 m to 41 m, within a meter of its shallowest, 40.9 m, from day 72 to day 137, and beneath it a seasonal thermocline forms. On day 120 the water is 16.3 °C at 40 m and 12.3 °C at 60 m. The surface temperature rises to 16.7 °C on day 147, 56 days after the strongest heating: a seasonal lag. The slab, with about a quarter of the mixed layer's heat capacity, peaks sooner and higher, at 23.6 °C on day 129.

Yet the sea stores far more heat. On day 167 the top 100 m of the column holds 1.05 GJ/m² more than in March, enough to warm the whole air column above it by 103 K, while the slab stores at most 0.45 GJ/m², because its warm surface gives so much of the heating back.

From day 168 the column loses more heat than it gains. Cooled water at the surface is denser than the water just beneath it, so it sinks and keeps the water stirred, the overturning of Part 1 driven from the top, which in the sea is called convective mixing. The energy it releases joins the wind's in stirring the boundary layer, and the mixed layer deepens from 61 m on day 200 to 90 m on day 250, 122 m on day 300 and 151 m on day 350, about 0.6 m a day. Meanwhile the surface cools from its peak only to 10.7 °C on day 355, because each loss of heat is shared with all the water it has taken in: 151 m of sea holds the heat of 60 air columns for each kelvin. The boundary layer, the depth the wind and the convection are stirring now, is deepest on day 347, at 155 m. In the last weeks of the year the heating wins again and the boundary layer retreats, to 86 m on day 365, but the mixed layer stays at 150 m, because the water below the boundary layer is still uniform and will be until the next spring's warmth stratifies it.

Over the year the surface swings by 6.0 K and the slab's by 20.8 K, within the 20 to 26 K that Part 7's slab swung near 45°N. Part 7's slab was warmest three months after the June solstice, though, and the figure's slab only 38 days after it: the figure takes back 10 W/m² for every kelvin the surface warms, and so strong a pull keeps a shallow sea in step with its heating. Land keeps its seasons' heat in its top few meters and swings further still, which is why, as Part 1 put it, land warms faster than the sea beside it.

Set the wind stress to 0, and nothing stirs the warmth down: the mixed layer stays in the top layers, under 6 m, from day 20 to day 133, the surface warms to 20.3 °C, 23 days after the strongest heating, and swings by 10.3 K over the year, and the top 100 m stores only 0.60 GJ/m². Convection alone still mixes the sea down to 161 m by winter, and the mixed layer is gone by day 360. Slide the latitude to 15°N instead, where the heating swings by only ±49 W/m², and the surface swings by 2.2 K and the slab's by 7.7 K; with no wind the surface swings by 4.6 K.

The faint lines in the year are the interfaces between the layers, and the bar at the left edge of the column shows each layer's mode: pale where the layer is held at its minimum thickness, bright where it follows its isopycnals. At 45°N the first 40 layers, down to 334 m, keep their minimum thicknesses all year. The start's water is denser than most of their label densities, so they stack from the surface at those thicknesses and hold whatever water lies in them; they are fixed-depth. Only below 334 m are layers thicker than their minimum, 62 to 148 m, and follow surfaces of equal density; they are isopycnal. At 15°N the isopycnal layers begin at 264 m. The deepest mixed layer of the year, 151 m at 45°N and 119 m at 15°N, does not reach them, so the lines stay flat.

Two of the model's simplifications show here. The figure has no day and night: it takes the heating as a smooth curve through the year, where a day of sunlight averaging 250 W/m² would warm the top layer by 1.2 K, with the share it keeps, if nothing mixed it, and the night's convection would mix that warmth down. The model can add a diurnal warm layer and a cool skin to its sea, but they are off unless asked for, and the figure has neither. And the mixed layer here is whatever the wind's work and the cooling can mix. At 15°N it is 54 m at its shallowest, at the deep end of the 20 to 60 m that a climatology of 2004 gives much of the tropical ocean; in the Southern Ocean's winter the real one reaches 100 to 400 m.

What the model computes

The ocean takes a step of its own every few of the atmosphere's, the “longer step of its own” of Part 4: every four steps of the air on grids coarser than N = 64, which is every 1,800 seconds at Part 7's N = 16; from N = 64 up it is a set interval of model time instead, 22.5 minutes at N = 128. Between its steps the atmosphere's surface update warms or cools the open sea's surface temperature T_s as it would a slab's, at the heat capacity of the ocean's top layer, h_0 thick, that the ocean last handed it:

C = \rho_0\,c_p\,\max(h_0,\ 1\ \text{m}), \qquad \rho_0\,c_p = 1025 \times 3985 = 4.08 \times 10^{6}\ \text{J/m}^3\text{/K}

With h_0 = 3 m that is 12.3 MJ/m²/K. The top layer keeps the sunlight it absorbs, and the update sets the rest aside for the ocean to deposit below: the share that passes the layer, by Jerlov's two bands for clear ocean water, type IB. The share of the absorbed sunlight that passes a depth z is

\tau(z) = 0.67\,e^{-z/(1\ \text{m})} + 0.33\,e^{-z/(17\ \text{m})}, \qquad \tau(3\ \text{m}) = 0.31, \quad \tau(50\ \text{m}) = 0.017

less its value at 300 m, below which none passes. At its own step the ocean takes T_s as the temperature of its top layer, T_0, and deposits the light set aside over the layers beneath, each taking the fraction \left[\tau(z_k) - \tau(z_{k+1})\right]/\tau(h_0) of it, with z_k the depth of the top of layer k. Then it moves its water by the rules of Part 4, regrids, mixes each column, salts the top layer with the step's evaporation, rain and ice, and hands its new T_0 back. Apart from the sunlight set aside for the layers below, it never sees a flux of heat, only what the flux did to the temperature. The figure does the same every 3 hours, with its prescribed heating, all of it counted as sunlight, in place of the atmosphere.

The regrid and the remap, which the next section describes, put each layer where its water follows its isopycnals or, where they lie closer than its minimum thickness, hold it at that minimum with whatever water lies there, and they conserve the water, its heat and its salt.

Then the ocean mixes each column, in this order. First comes convective adjustment: wherever a layer is denser than the one beneath it, the two take one temperature and one salinity, the means of their heat and salt, and the merging goes on down until the column is stable. The potential energy that the merged layers release, E_c for each square meter, if they include the top layer, goes to the next stage. Second comes the energetic boundary layer, after Reichl and Hallberg of 2018, with the sources of Kraus and Turner. The wind and the convection supply an energy for mixing during the step \Delta t,

E = \rho_0\,m_*\,u_*^3\,\Delta t + n_*\,E_c, \qquad m_* = \frac{1.25\,\Lambda}{1 + \left(h\,|f|\,/\,(0.5\,u_*)\right)^2}, \qquad n_* = 0.2

where u_*^2 = |\tau|/\rho_0 is the friction velocity of the wind's stress, \Lambda is the factor, up to 2, that the Langmuir circulation adds after Qing Li and his colleagues of 2017, and |f| is the Coriolis parameter, which limits most of the stirring to depths less than 0.5\,u_*/|f|, 48 m at 45°N under this wind. The mixing is done by a diffusivity that is zero at the surface and at the base of the boundary layer, of depth h,

\kappa(z) = 0.4\,v_*\,z\left(1 - \frac{z}{h}\right)^2, \qquad v_*^3 = u_*^3 + \frac{E_c}{\rho_0\,\Delta t}

with 0.4 the von Kármán constant of Part 7's evaporation. The ocean finds h, between a tenth of a meter and the whole column, as the depth at which the work of mixing the column with this diffusivity, the potential energy that mixing its heat and salt raises, equals E. Where the energy cannot mix even a tenth of a meter, as under no wind while the sea warms, h is zero.

Third, the whole column also mixes where it shears. The gradient Richardson number Ri, the stratification against the square of the shear between neighboring layers, sets the viscosity and diffusivity of Pacanowski and Philander of 1981:

\nu = \frac{\nu_0}{(1 + 5\,\text{Ri})^2} + \nu_b, \qquad \kappa = \frac{\nu}{1 + 5\,\text{Ri}} + \kappa_b

with \nu_0 = 10⁻² m²/s, \nu_b = 10⁻⁴ m²/s and \kappa_b = 10⁻⁵ m²/s. Where the column is still unstable, \kappa is 0.1 m²/s. In the figure the currents are set back to rest after every step, so the shear is only what one step of the wind raises in the top layer. Last, one implicit solve per column mixes the temperature and the salinity with the sum of the two diffusivities, which keeps both exactly, and a second on each edge, with the viscosity \nu plus the boundary layer's diffusivity, hands down the column the momentum the wind's stress put into the top layer.

The ocean measures the mixed layer's depth after the mixing as the depth at which the density first exceeds the top layer's by 0.03 kg/m³, interpolated between the centers of the layers, and the figure draws it. The boundary layer's depth is the depth the energy is mixing now, and the mixed layer's is the depth the water is uniform, which is why after the heating returns the second can be the deeper.

Seawater

Seawater holds about 35 grams of salt in every kilogram. That is its salinity, given here in practical salinity units, psu, which for seawater are close to grams per kilogram. Salt makes water heavier and warmth makes it lighter, and heavier water sinks beneath lighter, so the sea is stacked by density. The model computes the density of seawater from the simplified equation of state that Fabien Roquet and his colleagues published in 2015, with the coefficients of the NEMO ocean model. The figure draws it over the ocean's temperatures and salinities, with lines along which the density is the same, the sea's isopycnals.

The probe opens on water at 25 °C and 35 psu, like the surface of a subtropical sea, at 1022.41 kg/m³. Real seawater of that temperature and salinity weighs 1023.3 kg/m³: the model's formula makes every water about 0.9 kg/m³ lighter, which does no harm, because only differences in density push the water. A kelvin of warming lightens it by 0.313 kg/m³, its thermal expansion, and a psu of salt makes it heavier by 0.729 kg/m³, so there a psu is worth 2.3 K. Drag the probe down to −1.8 °C at 34 psu, like a polar sea in winter, at 1026.47 kg/m³. There a kelvin changes the density by only 0.047 kg/m³, so in the model's formula water near 25 °C expands 6.7 times as much for each kelvin as at its freezing point, while a psu still adds 0.795 kg/m³: near freezing a psu is worth 17 K.

So heat layers the warm sea and salt layers the cold one. In the tropics warm water floats on cooler across the thermocline, the depths where the temperature falls fastest; in the polar seas fresher water floats on saltier across a halocline. Fresh water is densest at 4 °C, but in the model's formula water of 35 psu would be densest at −6.8 °C, so seawater grows denser all the way down to its freezing point, and a cooling sea keeps overturning until its surface freezes.

Because the lines curve, mixing makes water denser. Choose “mix two waters”: water at 0 °C and 34.54 psu and water at 15 °C and 37.33 psu have the same density, 1026.800 kg/m³, but half and half they make water at 7.50 °C and 35.93 psu that is 0.303 kg/m³ denser than either, heavy enough to sink beneath both. This is cabbeling. It comes mostly from the way the thermal expansion grows with temperature: the warm water gains more density by cooling 7.5 K than the cold water loses by warming 7.5 K.

The faint lines are the model's own layers. Each of its 58 layers has a label, a density, and a layer that follows its isopycnals holds the water whose density is nearest its label: the yellow band, which follows the probe, is the water of the probe’s nearest label. The labels are packed where the ocean holds its water, 0.25 kg/m³ apart down the thermocline from 1022.0 to 1026.0, 0.1 below it to 1026.6, and only 0.03 from 1026.65 to 1026.98, where the deep waters differ by little. There North Atlantic Deep Water and Antarctic Bottom Water, marked on the plane, fall in separate layers: at 2.5 °C and 34.92 psu the first weighs 1026.901 kg/m³ and belongs to the 1026.89 layer, and at −0.5 °C and 34.66 psu the second weighs 1026.926 kg/m³ and belongs to the 1026.92 layer. The mix of the last paragraph, 1027.103 kg/m³, belongs to the 1027.1 layer, seven layers below the 1026.8 layer of the two waters. Above 1022.0 come nine labels for the warm pool, and above those twelve more, 0.25 apart from 1016.5 to 1019.25 and dotted, lighter than any water of the start in the next view, whose lightest weighs 1021.03 kg/m³.

The layers are the ocean's counterpart of the σ layers of Part 3. Below the reach of the sunlight and the winds little makes water denser or lighter, so the sea's water moves along its isopycnals, as air that is neither heated nor cooled rides the isentropes of Part 4, and a layer that follows its isopycnals holds one water wherever it goes. But near the surface, and wherever the water is nearly uniform, isopycnals crowd together or leave the column, and a layer that followed them would shrink to nothing. So each layer also has a minimum thickness, and where its isopycnals lie closer than that the layer is held at its minimum and holds whatever water lies there: it is fixed-depth. The twelve lightest layers always are, the 57 m of surface layers of the first section. Below them the minimum is 8 to 10 m for the nine of the warm pool, 10 m for the 23 of the thermocline and the intermediate water, and 40 m growing to 500 m for the 14 deep ones. This is the hybrid coordinate of Bleck of 2002, with the change from fixed-depth to isopycnal made layer by layer, wherever the minimum is met.

Switch to the layers by latitude, the ocean the model builds when it has no atlas to start from: a sea 4,000 m deep with no land, under the climate model's opening sea surface, 28 °C on the equator, after the first regrid. Dotted interfaces bound fixed-depth layers and solid ones isopycnal layers. At the equator 28 layers are fixed-depth, the twelve at the surface, the nine of the warm pool and seven more, down to 214 m. Then 21 are isopycnal, from the 1023.75 layer, 22 m thick, to the 1026.77 layer, which holds the start's uniform deep water, 2.07 °C and 34.70 psu, from 2,519 to 3,997 m. The nine densest layers have no water and lie as sheets 0.5 m thick on the floor.

The probe opens on the equator in the 1026.2 layer, isopycnal, 837 to 912 m down at 6.84 °C. Follow it poleward. At 30.3°N it lies 460 m higher, from 377 to 430 m, at 9.63 °C: like the isentropes, the isopycnals climb toward the poles. But a layer cannot lie shallower than the minima of the layers above it add up to, 324 m for this one. It reaches that depth at 37.0°N, and with its top held there its base keeps rising, until at 43.9°N it is down to its minimum of 10 m. Poleward of there it is fixed-depth and holds whatever lies at 324 m: at 60.1°N water at 2.55 °C and 34.47 psu that weighs 1026.54 kg/m³, 0.34 more than its label, while the water of the 1026.2 label has risen to the surface, which weighs 1026.17. Fewer layers follow the more nearly uniform water poleward: 21 are isopycnal at the equator, 12 at 30.3°N, 6 at 60.1°N and 2 at 79.9°N.

What the model computes

The density depends on temperature and salinity alone. It is the density each water would have at the pressure of the surface, its potential density. Real water is squeezed by its own weight, nearly 2 percent denser 4 km down, but the squeezing changes which of two waters is the heavier only for waters of nearly the same density, so the model compares all of them at the surface. With T_a the temperature less 10 °C and S_a the salinity less 35 psu, the formula is

\rho = 1026 - a_0\left(1 + \tfrac{1}{2}\lambda_1 T_a\right)T_a + b_0\left(1 - \tfrac{1}{2}\lambda_2 S_a\right)S_a - \nu\,T_a S_a

with a_0 = 0.1655 kg/m³ per K, b_0 = 0.76554 kg/m³ per psu, \lambda_1 = 0.05952 per K, \lambda_2 = 7.4914 × 10⁻⁴ per psu and \nu = 2.4341 × 10⁻³ kg/m³ per K per psu. The 1026 kg/m³ it starts from is its density at 10 °C and 35 psu, 0.95 kg/m³ less than real seawater's, and from −1.8 to 30 °C and from 33 to 37 psu it stays 0.87 to 1.08 kg/m³ below the real density. Its thermal expansion and haline contraction, which the probe reads, are

\begin{aligned} \alpha &= -\frac{1}{1026}\frac{\partial \rho}{\partial T} = \frac{a_0\,(1 + \lambda_1 T_a) + \nu\,S_a}{1026} \\ \beta &= \frac{1}{1026}\frac{\partial \rho}{\partial S} = \frac{b_0\,(1 - \lambda_2 S_a) - \nu\,T_a}{1026} \end{aligned}

The term in \lambda_1 makes \alpha grow with temperature, by about 6 percent of its value at 10 °C for every kelvin. When two waters mix half and half, the terms that are straight lines average away, and what is left is the cabbeling:

\rho\left(\bar{T}, \bar{S}\right) - \tfrac{1}{2}\left(\rho_1 + \rho_2\right) = \tfrac{1}{8}\left(a_0\lambda_1\,\Delta T^2 + 2\nu\,\Delta T\,\Delta S + b_0\lambda_2\,\Delta S^2\right)

where \Delta T and \Delta S are the differences between the two waters. The formula has no term for pressure, so it also misses the way cold water is squeezed more than warm, which matters most for the deep waters. In real seawater, water like North Atlantic Deep Water, at 2.5 °C and 34.92 psu, and water like Antarctic Bottom Water, at −0.5 °C and 34.66 psu, are almost equally dense at the surface, the warmer heavier by 0.01 kg/m³; 4 km down the colder is heavier by 0.33 kg/m³, which is why it lies beneath. The model's formula puts the colder heavier already at the surface, by 0.024 kg/m³, which orders its deep layers rightly for the formula's reason and not the sea's. Near freezing the formula makes water expand nearly twice as much for each kelvin as real seawater, 0.46 × 10⁻⁴ against 0.25 × 10⁻⁴ per kelvin at −1.8 °C and 34 psu, so real polar water is layered by its salt even more than the model's. And it is poor for fresh water: at 0 psu it puts the density maximum at 1.85 °C instead of 3.98, which is why the plane starts at 30 psu.

After every step of the ocean the model regrids each column. Number the interfaces from the surface, so that layer k lies between interfaces k and k+1 and has a label \rho_k, a minimum thickness \Delta z_k and a thickness h_k. For each interface the regrid finds the depth d_k where the column's density, held from decreasing with depth and drawn as a parabola through each layer, crosses the density half-way between the labels on either side, \tfrac{1}{2}(\rho_{k-1} + \rho_k): the isopycnal between one layer's water and the next's. The new interface is that depth or the layer above held at its minimum, whichever is deeper,

\hat z_{k+1} = \max\left(d_{k+1},\ \hat z_k + \Delta z_k\right), \qquad \hat z_0 = 0

except for the twelve surface layers, which always take \hat z_k + \Delta z_k. A layer whose isopycnals lie closer than its minimum is held at it, and otherwise follows them. Because each interface is measured from the one above, no layer lies shallower than the minima above it add up to, except where the floor squeezes the column. Every layer keeps at least h_{\min} = 0.5 m, so a layer with no water is a sheet and not a gap. The mode weight

w_k = \min\left(1,\ \max\left(0,\ \frac{h_k - \Delta z_k}{\delta}\right)\right), \qquad \delta = 5\ \text{m}

is 0 for a fixed-depth layer and 1 for one at least 5 m thicker than its minimum. The eddy transport of the interfaces reads it, so that fixed-depth interfaces carry nothing, and so does the next regrid, which flattens a layer's parabola by the weights of its two interfaces, so that a column of uniform layers at their labels is regridded to itself.

The remap puts the water on the new layers. In each old layer the temperature, and likewise the salinity, is a parabola in \xi = (z - z_k)/h_k, as in Colella and Woodward of 1984,

T(\xi) = T_L + \xi\left[\left(T_R - T_L\right) + 6\left(\bar T - \tfrac{1}{2}(T_L + T_R)\right)(1 - \xi)\right]

with mean \bar T. The interface values T_L and T_R come from the quartic through the running integrals of the four layers around each interface and are held between the layer means they separate, and a parabola that would overshoot them is reshaped, so the remap makes no new extreme. A new layer's heat and salt are the integrals of the parabolas over its new interfaces,

(h\,T)_k^{\text{new}} = \int_{\hat z_k}^{\hat z_{k+1}} T\,dz

so the column's water, heat and salt are conserved to rounding. Currents are remapped on each edge in the same conservative way, but as straight lines with a limited slope, and the transport through the edge is put back as it was.

Pushed by the wind

Wind blowing over the sea drags on it with its stress, a force on each square meter of sea. The model works it out in every cell from the wind of its lowest layer of air, about 20 m up, by the similarity theory of Part 7's evaporation: a wind of 8.9 m/s in neutral air pushes with 0.1 N/m², and one of 13.5 m/s three times as hard. The push goes into the sea's top layer, 3 m thick, and turbulence passes it down from layer to layer, and the layers it reaches move at different speeds.

Pushed steadily, the water does not keep moving along the push. Part 2's parcel, pushed steadily by the pressure, ended up drifting at right angles to it, and the sea does the same under the wind, to the right of the wind in the northern hemisphere and to the left in the southern. Vagn Walfrid Ekman worked this out in 1902, to explain why Fridtjof Nansen had seen the Arctic's sea ice drift 20 to 40° to the right of the wind, and the drift is named after him, the Ekman transport.

The figure runs the model's own ocean on the N = 8 grid, with its 58 layers, at rest until a band of wind about 20° wide starts to blow east along one latitude, and averages everything around that circle. The mixed layer is the depth the model measures, as in the first section. One panel shows the paths a float would take in the top layer, in the mixed layer and in the water well below, which the wind does not reach. Another shows the velocity of the top layer and of the mixed layer against that water, their tips tracing hodographs, with the two forces on the mixed layer, the wind's push and the Coriolis force. The third shows the current in each layer beside Ekman's spiral.

At 45°N under 0.1 N/m², the wind first pushes the top layer east, and the Coriolis force turns it to the right. The mixed layer starts 5 m deep, and the wind stirs it down to 36 m in four days. Its velocity against the water well below swings around a circle once every 16.9 hours, Part 2's inertial circle seen as a velocity, at up to 9.2 cm/s, and the loops die away only slowly: on day 4 they still reach 1 cm/s around a steady drift of 2.6 cm/s, 87° to the right of the wind, which is why the readouts average over two whole circles. The top layer, where the wind's push enters, drifts faster, at 5.5 cm/s and 53° to the right. The two forces on the mixed layer then nearly cancel: the Coriolis force holds the layer against the wind, as it held Part 2's parcel against the pressure.

That balance sets how much water the drift carries. The Coriolis force on a layer is f times its speed times its thickness, and the layers together must match the wind's push, τ/ρ₀, so the speed times the depth, added up through the water the wind moves, is τ/(ρ₀f). The figure adds up each layer's current against the water two to three boundary-layer depths down. The readout's 0.893 m²/s is 0.99 times the 0.903 m²/s that τ/(ρ₀f) gives under the band of wind, 88° to the right of the wind, and the mixed layer carries nearly all of it. Toward the equator f shrinks and the balance weakens: at 15°N the water carries 1.08 times τ/(ρ₀f) at 95° to the right of the wind. On the N = 8 grid the cells averaged for 5°N already reach from 1° to 8°N, and there the drift turns only 66° from the wind and carries 0.86 of τ/(ρ₀f); on the equator f vanishes and there is no Ekman balance at all.

Ekman's theory passes the wind's momentum down from depth to depth by turbulence, an eddy viscosity, and the Coriolis force turns each depth a little further, so the current turns and weakens downward in Ekman's spiral. The model does the same, and the third panel draws both: the dots are the model's layers, and the dashed spiral is the theory at the one constant viscosity that fits them best, 0.018 m²/s, which is theory, not the model. At 45°N the model's current turns from 53° in the top layer to 94° at 16 m and 145° at 29 m, where it has weakened to 1.6 cm/s, and by 40 m it is gone. The spiral reaches 6.8 cm/s at the surface, runs the opposite way at 59 m and matches the layers to within 0.45 cm/s.

The match is not exact, because the model's viscosity is not constant: it peaks at 0.025 m²/s at 8 m, falls to 0.003 at 29 m and to 10⁻⁴ below the boundary layer, and it changes in time as the boundary layer deepens, from 21 m after 9 hours to 39 m after four days. Toward the equator the departures grow: the best constant viscosity misses the model's layers by 14 percent of its surface current at 15°N and 20 percent at 5°N, against 7 percent at 45°N.

Switch to the whole planet, with the trade winds blowing west within 30° of the equator and the westerlies east from 30° to 65°. Under the trades the drift runs away from the equator on both sides, and water rises from below to replace it, 0.92 m a day on the equator at a depth of 100 m: upwelling. Away from the equator f grows and the trades weaken, so their drift slows as it goes and the water it brings piles up and sinks: set the latitude to 15°N, and there it sinks 0.36 m a day. Under the westerlies the drift runs toward the equator, carrying 0.96 and 0.93 times τ/(ρ₀f) at 42.5° and 47.5°, and near 30°, where it meets the trades' drift, the water sinks again, 0.16 m a day. The sinking is patchy on the N = 8 grid, and at 25° the water even rises a little, 0.09 m a day, but it is this sinking that drives the gyres of the section after next.

What the model computes

The wind's stress enters the top layer as a force spread through its thickness h_0, about 3 m. On each edge, the top layer's velocity \mathbf{u}_0 changes by

\frac{\partial \mathbf{u}_0}{\partial t} = \class{term-a}{-f\,\hat{\mathbf{z}}\times\mathbf{u}_0} \class{term-e}{+ \frac{\boldsymbol{\tau}}{\rho_0\,h_e}} + \cdots

where h_e is the layer's thickness at the edge. The dots stand for the pressure, the turning by the flow's own spin and the smoothing, which the next section takes up. At the end of each step the model solves for the velocity of every layer on the edge at once, with a viscosity \nu between each pair of layers. A layer gives up the momentum the layer beneath takes, so the solve hands momentum down without making or losing any:

h_k\,\frac{u_k^{\text{new}} - u_k}{\Delta t} = \class{term-e}{\nu_{k+\frac12}}\,\frac{u_{k+1}^{\text{new}} - u_k^{\text{new}}}{d_{k+\frac12}} - \class{term-e}{\nu_{k-\frac12}}\,\frac{u_k^{\text{new}} - u_{k-1}^{\text{new}}}{d_{k-\frac12}}

with d the distance between the centers of the two layers. The viscosity is the sum of two parts. In the boundary layer, from the surface to the depth h that the energy budget of the first section sets, it is the diffusivity that mixes heat and salt there, 0.4\,v_*\,z\,(1 - z/h)^2 at a depth z, where v_* is the friction velocity u_* = \sqrt{|\boldsymbol{\tau}|/\rho_0} under the wind alone, as in the figure, 0.99 cm/s under 0.1 N/m². Throughout the column the interior adds \nu_0/(1 + 5\,\mathrm{Ri})^2 + \nu_b, the first section's shear mixing. The turning planet caps the boundary layer near 0.5\,u_*/|f|, 48 m at 45° under 0.1 N/m², and in the figure it is 39 m deep after four days.

The coupled model hands the ocean the stress of its lowest layer of air, \rho\,C_D\,|\mathbf{U}|\,\mathbf{U}, once at each of the ocean's steps. Sea ice in the model does not move, and where it covers a share c of the sea, 80 percent of the stress passes through it to the water:

\boldsymbol{\tau}_{\text{sea}} = (1 - 0.2\,c)\,\boldsymbol{\tau}

Ekman's theory, which the model does not use, balances the Coriolis force at each depth against a constant eddy viscosity's drag between depths, with the wind's stress at the surface:

\class{term-a}{f\,\hat{\mathbf{z}}\times\mathbf{u}} = \class{term-e}{A\,\frac{\partial^2 \mathbf{u}}{\partial z^2}}, \qquad \rho_0\,A\,\frac{\partial \mathbf{u}}{\partial z} = \boldsymbol{\tau}\ \text{at}\ z = 0

For a wind of stress τ blowing east in the northern hemisphere, with z negative downward, its solution is the spiral

\begin{aligned} u &= V_0\,e^{z/d}\cos\left(\frac{z}{d} - \frac{\pi}{4}\right), \qquad v = V_0\,e^{z/d}\sin\left(\frac{z}{d} - \frac{\pi}{4}\right) \\ d &= \sqrt{\frac{2A}{f}}, \qquad V_0 = \frac{\tau}{\rho_0\,\sqrt{A\,f}} \end{aligned}

Added up over all depths, u comes to zero and v to −τ/(ρ₀f), the transport to the right of the wind. The model's balance is the same with A replaced by the viscosity above, and when the flow is steady and no pressure force acts on it, the stress the layers pass between them cancels when they are added up and only the wind's enters. So the transport comes out at 0.99 of τ/(ρ₀f) while the shape of the spiral differs from the theory's: the transport needs only the wind and f, the shape needs the viscosity. The figure's spiral takes for A the constant whose spiral is closest to the model's currents, by least squares over the layers above twice the boundary layer's depth.

On Earth the wind's momentum reaches an Ekman layer about 0.4\,u_*/|f| deep, 38 m at 45° under 0.1 N/m². In the model it reaches as far as the boundary layer does, about 37 m at 45°N, 40 m at 15°N and 50 m at 5°N in the figure's last two circles, and the layers below feel the wind only through its curl. Where the drift converges, the layers beneath it are pressed down and the sea surface rises, and where it diverges they rise, as the gyres need.

A sloping sea

In Part 2 the wind ended up blowing along the lines of equal pressure, held there by the Coriolis force. The sea does the same, and its pressure is simpler to find: at any depth it is the weight of the water above, set by the height of the sea surface and the density of every layer between. Where water piles up, the pressure beneath it is higher than around it, and the current that balances the outward push runs around the pile instead, a geostrophic current.

The figure runs the model's own ocean with its 58 layers on the N = 16 grid, over a flat floor 4,000 m deep, with no wind and no heating. The water is 27 °C at the surface and cools with depth. The top 33 layers, down to 264 m, are fixed in depth, from 3 m thick at the top to 10 m; below them each layer holds water of one density and follows it up and down, from 1025.0 kg/m³ to the bottom water at 1026.77. The 1025.25 layer is 46 m thick and lies 288 to 334 m down. At the start it is thicker over a patch about 3,000 km across near 45°N, by 60 m at its center, with the same thickness taken from the bottom water, so every layer between them is pushed down by 60 m and the column holds the same water. The fixed-depth layers have not changed, and the sea surface is flat.

The column under the lens is lighter than the columns around it, so deep down the pressure there is lower, and the deep water flows in from every side. The sea surface at the center rises, overshoots and rings for a few days, and on day 10 it stands 7.6 cm high over a lens base nearly 60 m deeper. Had the pressure on the bottom stayed what it was, its rise would be the lens's thickness times the difference in density between the bottom water and the lens layer, 1.5 kg/m³ out of 1025, which comes to 8.8 cm. The bottom pressure is 1.2 cm of water lower under the lens: the lens took mass out of the column, and the gravity waves of the first hours that carry the loss away leave 13 percent of it behind on day 10, close to the 16 percent that the bump below leaves at its center.

The top layer's current circles the pile clockwise, as Part 2's air circled a high, at 0.35 cm/s on day 10, and the Coriolis force on it matches the slope of the sea surface to within a percent, the readout's 0.99 times the geostrophic current. All 33 fixed-depth layers move alike, because the sea surface's slope is the only pressure that changes along them. Below, the currents fade: the lens layer moves at 0.34 cm/s, the fifth layer beneath it at 0.09 and the bottom water at only 0.06. At their depth the two pushes cancel, the extra height of the surface adding weight to every column near the center and the lens's lighter water taking the same weight away. That is Part 5's thermal wind in water.

Now make a bump in the sea surface instead, as Part 5 dropped its stone, by making the bottom water 1 m thicker over the same patch. Within three hours it has collapsed into a ring of gravity waves, whose leading edge the readout times at 186 m/s, close to the 198 m/s of a wave in water 4,000 m deep. An eddy 15.8 cm high is left, circled by every layer alike, and after a day only 31 percent of the raised water is still within 3,000 km: the radius Part 5 found, the wave's speed over f, is here about 1,900 km, close to the bump's own size.

The lens stays because its radius is far smaller. A wave along the stratification, the density rising by 5 kg/m³ through the top kilometer, travels at only 2.67 m/s, and divided by f that gives a deformation radius of 26 km at 45°, a hundredth of the lens's width. That radius is also the size of the eddies that the real ocean's currents shed, its mesoscale eddies. Even the climate model's N = 128 grid, with cells 60 km apart, cannot draw them, so the model's ocean makes none of its own, and an eddy transport flattens the slopes of the layers that follow density as real eddies would; here it lifts the lens's base by only 0.35 m in 30 days.

What the model computes

Each layer is a layer of water like Part 5's single layer, with its own thickness h_k at the cell centers and its own velocity across each edge, and the model steps both with Part 4's rules. A layer's thickness changes by what flows in and out through the walls:

\class{term-a}{\frac{\partial h_k}{\partial t}} = \class{term-b}{-\nabla\cdot(F_k)}, \qquad F_k = \min(h_e,\,h_{\text{donor}})\,u_k

Across an edge the top layer has the mean of its two cells' thicknesses and every layer beneath it the smaller of the two, h_e, and what is carried is no more than the thickness h_{\text{donor}} of the cell the water leaves, so a layer never draws water from a cell where it has none, and with the smaller thickness at the edge it carries next to nothing into one either. A whole column carries no more water across an edge than both sides hold, which matters at a coast or over a ridge. The dynamics move no water from one layer to another; after each step the model regrids every column, returning each interface to its layer's fixed depth or to where the density crosses the middle between two labels, and remaps heat and salt onto the new layers, conserving water, heat and salt. The velocity follows Part 4's momentum equation, with the same colors:

\frac{\partial \mathbf{u}_k}{\partial t} = \class{term-a}{-(\zeta_k + f)\,\hat{\mathbf{z}}\times\mathbf{u}_k} \class{term-b}{-\nabla K_k + \mathbf{P}_k} \class{term-e}{+\,\mathbf{F}_k}

The turning is Part 4's, computed the same way from the layer's potential vorticity (\zeta_k + f)/h_k and its flux h_k\mathbf{u}_k, and K_k = \tfrac{1}{2}|\mathbf{u}_k|^2. The forcing \mathbf{F}_k gathers the wind's stress on the top layer and the viscosity that hands it down, the last section's, a drag C_D|\mathbf{u}|\mathbf{u} on the water within 50 m of the floor, with C_D = 3\times 10^{-3}, and Part 4's smoothing -\nu_4\nabla^4\mathbf{u}, which damps the grid scale in 12 hours, with \nu_4 = 1.27\times 10^{16} m⁴/s at N = 16. The pressure force \mathbf{P}_k takes the place of Part 4's geopotential. For a layer at its label, whose water has the same density \rho_k everywhere, it is the gradient of the Montgomery potential, which holds the weight of the layers above it with their thicknesses h_j and densities \rho_j:

\class{term-b}{\Phi_k} = \class{term-b}{g\,\eta + \frac{g}{\rho_0}\sum_{j<k}(\rho_j - \rho_k)\,h_j}, \qquad \mathbf{P}_k = -\nabla\Phi_k

where \eta is the height of the sea surface and the density \rho_0 = 1025 kg/m³ stands in for each layer's own in the first term, by at most 0.3 percent in the figure's water, the Boussinesq approximation, which the model makes throughout. A fixed-depth layer holds water of changing density, so its density varies along it and the Montgomery potential does not apply. The model therefore takes, on every layer, the hydrostatic pressure force on an edge between cells a and b at one depth z, the middle of the depths the layer occupies in both cells, or of the gap between them where it occupies none in common, with H_k = \sum_{j<k} h_j the depth to the layer's top in each cell and d_c the distance between the cell centers:

P_k = \class{term-b}{-g\,\frac{\eta_b - \eta_a}{d_c} - \frac{g}{\rho_0}\,\frac{\left[\sum_{j<k}\rho_j h_j\right]_b - \left[\sum_{j<k}\rho_j h_j\right]_a}{d_c}} \class{term-c}{- \frac{g}{\rho_0}\,\frac{\left[\rho_k\,(z - H_k)\right]_b - \left[\rho_k\,(z - H_k)\right]_a}{d_c}}

For a layer at the same depths on both sides this is the layer's mean form, the weight of the layers above plus a second pressure term -\tfrac{g\,h_k}{2\rho_0}\nabla\rho_k, and for a layer at its label it reduces to the Montgomery gradient above. When the currents change slowly, the turning balances the push, and a layer at its label runs along the lines of equal \Phi_k:

\class{term-a}{f\,\hat{\mathbf{z}}\times\mathbf{u}_k} = \class{term-b}{-\nabla\Phi_k}

In every fixed-depth layer above the lens the only pressure that changes along the layer is the sea surface's, so each runs at g|\nabla\eta|/f: a slope of 1 cm in 100 km drives 0.95 cm/s at 45°. Under the lens the bottom water stays still only if its \Phi is the same everywhere. Of the layers above it only the lens layer changed its thickness, so

\begin{aligned} g\,\Delta\eta + \frac{g}{\rho_0}(\rho_{\text{lens}} - \rho_{\text{bottom}})\,\Delta h_{\text{lens}} &= 0 \\ \Delta\eta &= \frac{\rho_{\text{bottom}} - \rho_{\text{lens}}}{\rho_0}\,\Delta h_{\text{lens}} = \frac{1.515}{1025}\times 59.5\ \text{m} = 8.8\ \text{cm} \end{aligned}

which is the readout's mirror, less the 1.2 cm of bottom pressure left by the waves.

The eddy transport moves the interfaces that follow density toward the level, with a diffusivity \kappa of 1000 m²/s where the cells are far wider than the deformation radius, fading to nothing toward the surface above 200 m and toward the floor. Across an edge it carries a volume of water between layers in proportion to how much higher the interface lies on one side, weighted by how far the interface follows density, its mode weight w, the smaller of the two cells' values:

F_{k+\frac12} = -\kappa\,w_{k+\frac12}\,\frac{z_b - z_a}{d_c}, \qquad w_{k+\frac12} = \mathrm{clamp}\left(\frac{h_k - \Delta z_k}{5\ \text{m}},\,0,\,1\right)

where z_a and z_b are the depths of the interface in the two cells, d_c the distance between their centers and \Delta z_k the layer's minimum thickness. A layer held at its minimum thickness has weight 0, so the fixed-depth layers are not flattened; one thicker than its minimum by 5 m or more has weight 1.

The fastest wave in the sea runs along its surface, at 198 m/s over 4,000 m of water. Rather than take every step that short, the model splits it off. The layers take one RK4 step with the sea surface held still. Then the sea surface and the transport summed through the depth take short steps of their own, each no longer than

\Delta t_{\text{fast}} \le 0.35\,\frac{d_{\min}}{\sqrt{g\,D_{\max}}}

where d_{\min} is the shortest distance between two cell centers and D_{\max} the deepest water. At N = 16 the closest centers are 417 km apart, so the short step is at most 736 seconds, and the figure's ocean step of 3,600 seconds takes 5 of them. The layers are then stretched to the averaged sea surface and shifted to the averaged transport, and as a last resort any current faster than 5 m/s is cut back to 5 m/s, a cap that does not conserve momentum. With all 58 layers at N = 128 the ocean holds 57.0 million numbers, the thickness, heat and salt of every layer in every cell and its velocity across every edge, against the 24 million of the dry atmosphere in Part 4.

Walls make gyres

The trade winds of Part 5's dry sphere blew from the east at 15 to 20° of latitude and its westerlies around 45°, as the Earth's do, and two sections back their drifts met between them and pressed the water down. Under a wind stress that turns from 0.1 N/m² westward at 15° to as much eastward at 45°, the water near 30°N is pressed down by 11 cm a day.

On a spinning sphere that water drifts toward the equator. A column of water squashed from above grows wider and spins more slowly, as a skater does with her arms out, and nearly all of its spin is the planet's own, f, so it can spin more slowly only by moving to where f is smaller. Harald Sverdrup worked this out in 1947, and the relation between the wind and the drift is the Sverdrup balance. It depends on f growing toward the pole, the β that made Part 2's loops creep west.

The water comes back toward the pole in a narrow, fast current along the western side of the basin, like the Gulf Stream off North America and the Kuroshio off Japan. Such currents are measured in sverdrups, Sv, a million cubic meters of water a second, and the Gulf Stream carries about 30 Sv through the Straits of Florida. The figure runs the model's own ocean in a basin like the North Atlantic's subtropics, 80° of longitude wide, from 12° to 48°N and 4,000 m deep, with all 58 layers, 27 °C at the surface and cooling with depth, all at rest, under a wind that blows west over 15 to 30°N and east over 30 to 45°N. The bars under the map follow the water across 30°N.

Within days a current starts north along the western wall, and by day 20 it has settled. Averaged from the surface to the floor over 25 to 35°N, it runs north at 1.19 cm/s in a band 920 km wide that carries 37.7 Sv across 30°N, while the interior drifts south at 0.080 cm/s, 15 times slower. Sverdrup's balance asks for 0.116 cm/s and 35.7 Sv. Across the whole row 40.6 Sv flows north and 40.3 Sv back south, a little more than the balance asks, because some of the western current's water turns back south right beside it. The broad interior, still catching up with Sverdrup's balance, drifts south at about two-thirds of what the balance asks. Switch the arrows to the current averaged over the depth, and the whole basin turns clockwise, an ocean gyre like the Earth's in the northern subtropics; the top layer's current is mostly the wind's own Ekman drift, which converges on 30°N. By day 100 the western current carries 39.4 Sv and the interior drifts south at 0.096 cm/s, 12.9 times slower, and doubling the wind doubles every speed and transport.

Now make f the same everywhere, its value at 30°N, and Sverdrup's balance is gone: the water pressed down has no reason to drift south, and the western wall is no longer special. The gyre turns almost symmetric, north along the western wall at 1.83 cm/s on day 100 and south along the eastern at 1.35, and it never settles: only weak friction resists the spin the wind gives it, and it spins up from 75 Sv each way across 30°N on day 20 to 257 Sv on day 100. Henry Stommel showed in 1948 that the crowding against the western shore comes from the change of f with latitude.

Take the walls away instead, and there is no gyre at all: on day 100 only 1.5 Sv crosses 30°N northward. The water runs around the planet with the wind, west under the trades at 6.70 cm/s near 17.5°N and east under the westerlies at 5.79 cm/s near 37.5°N, still speeding up when the figure ends. The Earth's Southern Ocean is open like this, and there the westerlies drive the Antarctic Circumpolar Current all the way around Antarctica.

The model's western current is far broader and slower than the Gulf Stream, which is about 100 km wide and runs at up to 2.5 m/s: at N = 16 the model's cells are 481 km apart, and its current spreads over two of them and all 4,000 m of depth. Yet it carries about as much water, because Sverdrup's balance sets how much must come back from the wind alone. The 100 days show only the fast part of the gyre; the slow Rossby waves of Part 5 that would tilt the layers beneath it into a bowl, at a speed of β times the deformation radius squared, need about nine years to cross the basin at 30°N.

The figure's wind is prescribed rather than the model's atmosphere's, no heat or fresh water crosses its surface, and every column starts at rest in the same state, one profile of temperature and salinity with no tilt of the layers, over a flat floor, so the real ocean's slow adjustment of its thermocline to the winds, its continents' shapes and its seabed's ridges are left out.

What the model computes

Sverdrup's balance comes from the layer equations, added up from the floor to the surface. Far from the walls the flow is slow and broad, so its own spin, its momentum and friction all matter little, and the turning balances the pressure gradient and the wind. The curl of that balance removes the pressure, and since f changes with latitude, the turning leaves behind

\class{term-a}{\beta\,V} = \class{term-e}{\frac{1}{\rho_0}\,\hat{\mathbf{z}}\cdot\nabla\times\boldsymbol{\tau}}

where V is the northward transport per meter of width, the speed added up from the floor to the surface, \beta = \partial f/\partial y = 2\Omega\cos\varphi/a is how fast f grows northward, and \boldsymbol{\tau} is the wind stress. The figure's wind, \tau_x = -\tau_0\cos\left(\pi\,\frac{\varphi - 15^\circ}{30^\circ}\right) between 15° and 45°N, has a curl of -6\tau_0/a at 30°N, −9.42 × 10⁻⁸ N/m³ for \tau_0 = 0.1 N/m². There β is 1.98 × 10⁻¹¹ per meter per second, and V is −4.63 m²/s: 35.7 Sv south across the 7,700 km of the basin at 30°N, or 0.116 cm/s spread over 4,000 m. The water there sinks through the base of the Ekman layer at w = -\frac{1}{\rho_0 f}\,\frac{\partial \tau_x}{\partial y}, where the wind itself is zero: −1.26 × 10⁻⁶ m/s, the 11 cm a day of the first paragraph.

Stommel added a friction that grows with the current's speed and found the return squeezed against the western wall, in a layer whose width the friction sets. In the model that friction is mostly the smoothing of the currents, with \nu_4 = 1.27 × 10¹⁶ m⁴/s at N = 16. A current along a wall then has a natural width of about (\nu_4/\beta)^{1/5}, 230 km at 30°N, half a cell, so the grid sets the width instead: 920 km on day 20, from the coast to where the flow turns south. In a separate run with the smoothing four times weaker, the current narrows only to 690 km. At N = 128, with cells 60 km apart, the natural width is 66 km, close to the Gulf Stream's 100.

The frozen sea

At the model's freezing point, −1.8 °C, seawater freezes into ice that floats, and the ice changes the sea on both sides of it. It reflects sunlight that open water would absorb, Part 7's ice–albedo feedback. It slows the sea's loss of heat, which must now pass up through the ice. And it leaves most of its salt behind: the model's ice keeps 5 psu, about a seventh of the salt in water of 34 psu, and the rest goes into the water beneath, which is called brine rejection.

The figure runs the model's own ice code on one column at 75°N, over 50 m of water mixed through, a mixed layer of fixed depth chosen for this figure. The air above it is a toy written for this figure, not the model's: it takes more heat from a warmer surface, 200 W/m² at 0 °C and 2.1 W/m² more for each kelvin warmer, lets through 70 percent of the model's sunlight, and adds a fixed 85 W/m² for the heat the winds bring from warmer latitudes. The column starts as open water at 0 °C on the September equinox and runs for 20 years.

The water starts with 368 MJ/m² above freezing, and it takes 45 days to lose it. After that the water under the ice stays at −1.8 °C, and the sea loses heat only as fast as the ice conducts it up to the cold air. The ice's thermal conductivity is 2.0 W/m/K, so a meter of ice whose top is 20 K below freezing conducts 40 W/m², which the water pays for by freezing 1.1 cm of new ice a day onto its base. On the first December solstice the ice is 1.21 m thick over 90 percent of the sea, the rest open water between the floes, called leads. Thicker ice conducts less, so it grows ever more slowly, by the square-root law of the Stefan problem.

When the sun returns, the model's own ice reflects 62 percent of it while the ice is cold and 48 percent as it melts, and after its first summer it never melts away again, because each summer melts less than the winter before grew: in its twentieth year the sea holds 3.6 to 5.3 m of ice for each square meter, and more every year. Now choose Part 7's dark ice, which reflects 7 percent. It grows through the polar night exactly as the bright ice does, but it is gone 11 days before the June solstice, and the open water absorbs the summer's sunlight instead. By its tenth year the dark ice lasts only three months a year and is at most 0.94 m thick, over 85 percent of the sea, and the water under it warms to 5.3 °C in summer.

That is Part 7's feedback in its starkest form, since the toy air brings the same heat whatever lies below it, and no current brings any. Bring 60 W/m² instead, and the bright ice thickens without end, to 15.7 m after 20 years; bring 92, and even the bright ice melts away each summer, while the dark ice stops forming after four winters. Back at 85 W/m², turn the snow on: the toy's summer cannot melt snow that reflects 70 percent of the sunlight, so the ice beneath it thickens every year, bright or dark, to 12 m by the twentieth. Part 7's run darkened only its bare ice; its snow stayed bright until the summer melted it, which is why its feedback was weaker but not gone.

As the ice grows, the water beneath it grows saltier: every meter of ice frozen out of a 50 m mixed layer at 34 psu adds 0.53 psu to it and 0.42 kg/m³ to its density. The gauge sets that density against the model's deep water, the 2 °C and 34.7 psu that its ocean starts with at depth, 1026.77 kg/m³. Water at −1.8 °C reaches it at 34.38 psu, after 0.71 m of ice, and by the end of the first winter the column's water is 0.77 kg/m³ denser than the deep water. In the model such water sinks: its convective adjustment mixes it down through the layers beneath until it meets water as dense, and Antarctic Bottom Water forms this way, from the brine of the sea ice around Antarctica. The figure's column never mixes down, so under the bright ice its water grows saltier year after year, to 36.97 psu by the twentieth.

The deeper the mixed layer, the more summer heat it holds and the later it freezes. Over Part 7's 10 m slab the dark ice's open water warms to 22.0 °C yet holds only 1.00 GJ/m² above freezing, while the 50 m layer warms only to 5.3 °C but holds 1.46 GJ/m², enough to melt 4.8 m of ice, and under the dark ice a 200 m layer makes no ice at all after its third winter.

The figure leaves out the model's ocean around the ice: no current brings heat under it, and its mixed layer never mixes down. The model's ice itself never moves, and its leads open and close only by melting and freezing.

What the model computes

The model's sea ice follows the zero-layer model that Albert Semtner published in 1976, with a cover A: in each cell the ice is h thick over a share A of the sea, and the rest, the leads, is open water held at the freezing point T_f = 271.35 K. A skin at the top of the ice, with a small heat capacity C_s = 2 × 10⁵ J/m²/K, takes the net heat \class{term-e}{F} that the surface gains from the air: the sunlight it absorbs, its infrared, and its sensible and latent heat together. Heat conducts up through the ice and any snow on it, and the base freezes or melts by whatever of that heat the water beneath does not supply:

\begin{aligned} F_c &= \frac{T_f - T_s}{R}, \qquad R = \frac{\max(h,\ 0.1\ \text{m})}{k_i} + \frac{m_s}{\rho_s\,k_s} \\ C_s\,\frac{dT_s}{dt} &= \class{term-e}{F} + F_c \\ \rho_i L_f\,\frac{dh}{dt} &= F_c - F_o \end{aligned}

Here T_s is the skin's temperature, k_i = 2.0 W/m/K and k_s = 0.31 W/m/K are the conductivities of ice and snow, m_s is the snow's mass on each square meter of ice and \rho_s = 300 kg/m³ its density, and F_o is the heat the ocean hands to the ice's base. \rho_i L_f = 917 kg/m³ × 3.34 × 10⁵ J/kg = 3.06 × 10⁸ J/m³ is the latent heat of freezing a cubic meter of ice. A skin that would warm past 0 °C stays at 0 °C, and the extra heat melts first the snow and then the ice, from the top. The leads take their own share of the cell's heat. What they gain melts ice from the side, and the cover shrinks by half the volume's relative loss, as in the model that William Hibler published in 1979; what they lose freezes new ice, which closes them as if it were 0.3 m thick. The air's radiation sees the cell's albedo, its parts blended by area:

\begin{aligned} \alpha &= A\left[\alpha_b + (\alpha_n - \alpha_b)\min\left(1,\ \frac{m_s}{20\ \text{kg/m}^2}\right)\right] + (1 - A)\,\alpha_w \\ \alpha_b &= \alpha_w + \left(\alpha_i(T_s) - \alpha_w\right)\min\left(1,\ \frac{h}{0.5\ \text{m}}\right) \end{aligned}

\alpha_w is the open water's albedo, 0.02 under a high sun, 0.3 near the horizon and 0.06 for diffuse light. \alpha_i is the bare ice's, 0.62 while the ice is cold, falling over the last kelvin below melting to 0.48, which stands for melting ice with its ponds of meltwater; Part 7's dark ice set it to 0.07, cold or melting. \alpha_n is the snow's: fresh snow reflects 0.85, and near melting it closes about a fifth of its gap to 0.70 each day, until a fresh fall brings it back.

Under the ice lies the sea's top layer, not Part 7's slab: the ice's update takes the top layer's heat capacity, \rho_0 c_p\,\max(h_0,\ 1\ \text{m}) with h_0 = 3 m, in the slab's place. Wherever a cell has any ice, the ocean holds its top layer at the freezing point as each of its steps begins, and the wind's stress reaches the water at 0.8 of its strength under full ice cover. After the step has moved and mixed the water, so that warmer water from beneath may have reached the top layer, the ocean hands the heat the top layer holds above freezing to the ice's base, sets the layer back to freezing, and salts it by the change in the ice's volume, its cover times its thickness:

\begin{aligned} F_o &= \rho_0 c_p\,h_0\,\frac{T_0 - T_f}{\Delta t}, \qquad T_0 \leftarrow T_f \\ h_0\,\Delta S_0 &= (S_0 - 5)\,\frac{\rho_i}{1000\ \text{kg/m}^3}\,\Delta(A\,h) \end{aligned}

h_0, S_0 and T_0 are the top layer's thickness, salinity and temperature, T_0 as the step left it, and \Delta t is the ocean's step. The ratio \rho_i/1000 kg/m³ turns a volume of ice into the depth of fresh water it holds, and 5 psu is the ice's own salt. Melting runs the formula backward and freshens the top layer. The water's volume never changes, because the salt it gains or loses stands in for the fresh water the ice takes and gives back. The freezing point stays at −1.8 °C however salty the water is. The brine goes into a layer 3 m thick, so a centimeter of new ice salts it by 0.09 psu, enough to make it denser than the layer beneath, and at the next step the convective adjustment merges the two, and goes on merging down the column until the water below is as dense, so the brine reaches whatever depth its density does. The potential energy that merging releases joins the wind's in stirring the ocean's boundary layer.

A moving sea

The run below lets Part 7's sea move. It is Part 7's wet planet again, with the same air, grid and dark ice, but its slab of water 10 m deep is replaced by the model's own ocean, its 58 layers on a flat floor 4,000 m down. There is still no land, so the sea has no coasts. The sea started at rest from the model's own starting sea, as the section on seawater builds it, warm at the top and cooling to 2 °C below, and the run, on the model as it stood on October 18, 2026, went on for ten years and then recorded an eleventh, which begins at a March equinox.

The globe opens on the temperature of the sea's surface at the March equinox. Over the year it is warmest near the equator, 23.1 °C at 7.5°N and within 0.4 K of that from 12.5°S to 12.5°N, and coolest at the poles, 0.9 °C in the south and 1.7 °C in the north, and it hardly changes through the seasons: within 5° of the equator by 1.4 K, and near 45°N by 3.2 K, from 8.1 °C on March 30 to 11.3 °C on September 1, where Part 7's surface, partly frozen in winter, swung by 25.6 K.

Switch to the current: the top layer runs west under the trades, 90 and 86 cm/s in the bands centered 2.5° south and north of the equator over the year, and east under the westerlies, 10 to 12 cm/s near 40°. The fastest current anywhere in the top layer, in single cells, ran 1.2 to 1.8 m/s through the year. Switch to the upwelling: cell by cell it is a speckle at the scale of the grid, but averaged around the equator the water 50 m down rises 0.41 m a day at 2.5°S and 0.32 m a day at 2.5°N, sinks 0.31 m a day at 7.5° on both sides of the equator, and between 22.5° and 37.5° on either side sinks at 0.06 to 0.08 m a day.

Switch to the sea level. Over the year it stands highest under the subtropics, 90 cm above the mean at 22.5°S and 80 cm at 22.5°N, dips to 32 cm below the mean over the equator, and falls to 104 cm and 101 cm below it at 57.5°S and 62.5°N: nearly 2 m from hump to trough, the planet's ocean surface topography. The currents run along its slopes, as in the sloping sea: the fall of 38 cm between 32.5° and 37.5°N would balance the Coriolis force on a current of about 7.9 cm/s east, within a tenth of the 8.7 cm/s the top layer runs there.

Switch to the mixed-layer depth. It is shallowest on the equator, 18 m averaged around the circle, and from 12.5°S to 17.5°N it stays above the base of the twelve surface layers, 57 m. It deepens poleward, to 124 m at 37.5°S and 130 m at 42.5°N over the year. In winter the cooling surface mixes it down, as in the first section's column, to 380 m at 87.5°N in March and 356 m at 77.5°S in September, averaged around the latitude, and no single cell goes deeper than 485 m. Switch to the surface salinity: it is highest, 35.30 psu at 27.5°S and 35.27 psu at 27.5°N, a little poleward of where evaporation most exceeds the rain, at 22.5°, by 1.7 and 1.6 mm a day. Switch to the rain: in frames five days apart its band stays between 17.8°S in early January and 15.9°N in late September, where Part 7's ranged from 18.4°S to 23.3°N.

Show the first 60 days and switch to the current: on the equator the sea runs east at first, 20.2 cm/s on day 3 and 6.6 cm/s on day 25, turns west on day 32 and runs west at 16.6 cm/s on day 60.

The run's starting ice did not last: the north had ice for the last time on April 29 of the eighth year, and in the south a patch still forms around the pole each winter, 0.50 percent of its hemisphere at most, from July 22 to November 19 of the recorded year. The moving sea brings the polar seas heat that no current brought the frozen-sea column, 8.7 W/m² poleward of 60°N and 10.0 W/m² poleward of 60°S over the recorded year, which they give up to the air through their surfaces, with a little more from their store. Nearly free of ice, the planet is 3.1 K warmer than Part 7's, 288.5 K over the recorded year, and 18.1 K warmer at the South Pole and 8.8 K at the North Pole, while within 10° of the equator it is within 0.6 K of Part 7's.

The second figure averages the sea around each circle of latitude, layer by layer from the surface to the floor, with the overturning drawn as lines along which the water moves. It opens on the temperature over the year.

The warm water is a lens over the tropics and subtropics. The 1024.0 isopycnal, the surface of equal density that bounds it, lies 67 and 69 m down at 2.5°S and 2.5°N, 150 m at 17.5°S and 159 m at 17.5°N, and reaches the surface near 35°. Down to 284 m at 2.5°N every layer is held at its minimum thickness, and the first that follows its isopycnals is the 1025.5 layer; the layer labeled 1024.0, from 224 to 234 m, holds the 13.2 °C water that lies there, 1025.4 kg/m³, 1.4 more than its label. At 2.5°N, where the probe opens, the top layer is at 22.8 °C, the mixed layer is 19 m deep, and the deepest layer, from 1,991 to 3,995 m, is at 2.1 °C. Switch to the salinity: under the subtropics the water is saltier than 35 psu down to 735 to 784 m, and the polar seas are freshest at the top, 33.98 psu in the south and 34.11 psu in the north over deep water of 34.70 psu.

Switch to the eastward current, and the currents of the globe reach down to the floor. At 2.5°N the water runs west at 86 cm/s in the top layer, 70 cm/s at 57 m, 56 cm/s at 234 m, 27 cm/s at 748 m and still 9 cm/s in the deepest layer, 2,000 m thick: the whole column averages 18 to 19 cm/s west, and in the year's mean no layer near the equator runs east. As in the sea without walls of the section on gyres, there are no gyres, only belts that run around the planet, 2,412 Sv west between 30°S and 30°N and 320 and 275 Sv east under the westerlies. Nothing like a wall holds them back, and the run's records cannot show what balances the push of the winds instead.

The white lines follow the water. In the top 57 m the trades carry 107.3 Sv south across 5°S and 102.5 Sv north across 5°N, and these tropical cells are as deep as the lens: the flow above 300 m is still 96.2 Sv south and 91.5 Sv north, and it turns back between 300 and 500 m, so that the flow above 500 m is 7 Sv south on both sides. They rise on the equator, the upwelling of the globe, and sink near 7.5°.

Beneath them a deeper pair carries water poleward between 500 and 1,100 m, peaking at 33.8 Sv south across 10°S and 37.6 Sv north across 5°N, and returns it below, down to the floor; the run's records do not show what drives them. Under the westerlies two more cells turn the other way, 40.0 Sv north across 45°S and 36.6 Sv south across 40°N at their strongest, 351 and 289 m down; the top 57 m carry 31.4 Sv and 26.9 Sv across 45°S and 45°N, and by 1,000 m the flow is down to 1.6 and 3.7 Sv. Poleward of 60° no cell carries more than 7 Sv.

Choose September, and the southern cell grows and reaches across the equator: the top 57 m carry 237 Sv south across 5°S and only 46 Sv north across 5°N. Across the equator 34 Sv runs south in the top 57 m and 90 Sv above 300 m, and the flow south is still 68 Sv above 1,000 m. In March it is the northern cell that grows, 231 Sv north across 5°N against 32 Sv south across 5°S, and 111 Sv runs north across the equator above 100 m. Each hemisphere's cell grows in its winter, as Part 7's Hadley cell did, and over the year the flow across the equator never exceeds 15 Sv at any depth.

The densest layers lie as sheets. The nine of 1026.80 kg/m³ and heavier are 0.5 m thick on the floor, because the sea holds no water that dense: the deep water of its start is 1026.77 kg/m³, and the densest water at the surface in the recorded year, 1026.60 kg/m³, is lighter than the 1026.65 layer. Below 1,900 to 2,300 m in the tropics, and below 710 m at the poles, the sea is one layer, the 1026.77, at 2.0 to 2.1 °C and 34.70 psu, the water of its start.

The sea carries heat

On Part 7's planet the air carried all the heat that crossed each latitude, 4.2 PW north across 40°N and 5.0 PW south across 40°S. Now the sea carries some of it. A band of sea that takes in more heat through its surface than it stores hands the rest to its currents, so the sea's share can be added up across each latitude as Part 6 added up the planet's and Part 7 the air's. The figure shows the heat going into the sea and the heat it stored over the recorded year, and below them the heat carried across each latitude.

The sea takes in most of its heat on the equator, 50 W/m² in the two bands that border it, 2.4 PW within 5° of it and 3.6 PW between 10°S and 10°N. Show the surface's budget to see why. The sunlight it absorbs hardly changes across the tropics, 192 W/m² at 2.5°N and 174 W/m² at 17.5°N, and neither does the infrared it sends up, less what comes back, 45 W/m² at both. But on the equator the sea evaporates 3.0 mm of water a day and pays 88 W/m² for it, while at 17.5° it evaporates 4.0 mm, pays 115 W/m² and takes in only 1 to 2 W/m².

The dashed line is the heat the sea stored. Averaged over the planet the sea took in 8.5 W/m² and stored 8.5 W/m² over the year: the sea was still warming. Its surface had cooled from 16.2 °C at day 365 of the run to 15.2 °C at day 4015, while the mean temperature of the water below it rose from 3.87 to 4.02 °C, and it took in 4.9 W/m² in the second year and 8.5 W/m² in the recorded year. It stores most where the currents bring heat: 22.5 W/m² between 25° and 40°N and 25.0 W/m² between 25° and 40°S, where the surface takes in −1.1 W/m² in the north and 2.4 W/m² in the south, and it loses 17.4 W/m² within 5° of the equator, where the surface takes in 50, so the currents carry 68 W/m² away from the equatorial band. The polar seas lose 9.0 and 10.5 W/m² through their surfaces poleward of 60°, 0.3 and 0.5 of it from their store, and their currents bring the rest, 8.7 and 10.0.

The books close: the planet gained 8.43 W/m² at the top of the air and stored 8.44, 8.49 in the sea and −0.06 in the air, and the 0.003 W/m² the books cannot place is too little to move any curve of the figure.

Below, the planet carries 4.8 PW north across 30°N and 4.9 PW south across 30°S, corrected for the heat the sea stored. The air carries 3.9 PW at its peak, at 35° in each hemisphere, and 3.85 and 3.77 PW across 40°N and 40°S, where Part 7's air carried 4.2 and 5.0 PW; the sea carries the rest across 40°, 0.40 PW north and 0.43 PW south. From the top of the air alone the total would peak at 4.7 PW at 35°N and 4.6 PW at 35°S, 0.06 PW more than Part 7's from its top alone.

The planet absorbs 9.8 W/m² more sunlight than Part 7's, 4.3 W/m² of it from poleward of 50°, where the ice is gone, and 4.4 W/m² from within 30° of the equator, and it sends out as much infrared, 217.8 W/m² against 217.6, so it gains 8.4 W/m² at the top, where Part 7's planet lost 1.2. Its air holds 3 percent less vapor than Part 7's, 22.4 against 23.0 mm, and its vapor carries 1.9 PW north and 2.0 PW south across 35° against Part 7's 2.1 PW; its dry air carries 2.5 PW at 20° in each hemisphere.

The sea carries more than the air from 19°S to 19°N. It carries 2.4 PW north across 15°N and 2.3 PW south across 15°S at its peaks, and 74 and 72 percent of the total across 10°N and 10°S, 48 and 47 percent across 20°, 26 and 25 percent across 30°, 14 and 15 percent across 35°, and 9 and 10 percent across 40°, north first. No heat crosses the equator: across it the total is −0.01 PW, the sea's 0.03 and the air's −0.04, where Part 7's air carried 0.6 PW south.

Show the sea's overturning to see how. The sea moves heat in three ways: its currents carry it, its eddy transport flattens the slopes of its layers, and its lateral mixing spreads heat along them. Added up layer by layer, edge by edge, from the run's six-hourly states, the three carry across every latitude what the books say to within 0.03 PW. In the tropics the cells of the last section do most of the work. Across 10°S the layers send 97 Sv south and bring 97 Sv back: the water going south is 17.2 °C on average and the water coming back 10.2 °C, and the cells carry 2.8 PW south, where the sea's books say 2.2 PW: the eddy transport and the mixing carry 0.4 PW each back toward the equator. Across 5° the cells would carry 4.8 PW south and 4.7 PW north, but the sea's whole transport is 1.65 and 1.62 PW: the currents within each layer carry 2.2 and 2.1 PW of it back toward the equator, and the eddy transport and the mixing 0.9 PW more. The flow across a circle is far from uniform: the top 57 m move 272 Sv south and 235 Sv north across 10°S, at different edges and times, 37 Sv south in all. Across 20° the cells carry all of the 2.1 PW.

Across 35° and 40° the currents carry heat back toward the equator, 0.3 PW across 35°N and 0.4 PW across 40°N, in the cells under the westerlies. The sea's poleward heat there, 0.64 PW across 35°N and 0.40 PW across 40°N, is carried down the gradient of temperature: by the lateral mixing, 0.70 and 0.62 PW, and by the eddy transport, 0.23 and 0.19 PW. Most of the mixing, 0.48 and 0.44 PW, is in the layers below the twelve at the surface, where every layer takes the background diffusivity, 0.005 m/s times the spacing of the cells, 2,400 m²/s at N = 16. Both stand in for the eddies that a grid of cells 481 km apart cannot draw.

On the Earth the atmosphere carries 5.0 PW at its peak, near 43°N and about as much near 40°S. The total peaks near 35°, where the atmosphere carries 78 percent of it in the north and 92 percent in the south (Trenberth and Caron, 2001). This planet's sea carries 14 percent of the total across 35°N and 15 percent across 35°S, against the Earth's ocean's 22 and 8 percent. Much of the Earth's ocean heat travels in the narrow, fast boundary currents that walls make, the Gulf Stream among them. This sea has none, and carries its heat in its broad tropical cells and, farther out, by the mixing and the eddy transport that stand in for the eddies its grid cannot draw.

From here to the land

The model's sea stores the seasons' heat, moves its water along layers of equal density where it can, drifts across the wind, gathers into gyres against walls and freezes in the polar seas. Let go on Part 7's planet, it carried most of the heat that crosses 10° and a seventh of what crosses 35°, and its dark starting ice melted away; only a patch around the South Pole freezes again each winter.

Parts of it are not finished. This sea has not settled: in the recorded year it took in 8.5 W/m², and the water below its surface warmed from 3.999 to 4.015 °C. It has no coasts, so nothing holds back the winds' push: near the equator every layer runs west, 86 cm/s at the top and 9 cm/s in the deepest layer, where the real Pacific carries an eastward undercurrent of 0.5 to 1 m/s. The model's own design notes list settings still to be tested against the real sea: the thicknesses of the deepest layers against the stratification of the Southern Ocean and the Arctic, the rotational limit and the Langmuir cap of the boundary layer's mixing against measured mixed-layer depths, the background diffusivity and the surface taper of the eddy transport. The next part adds the land, whose coasts shape the real ocean's gyres.