How ENSO works

The resting state

The tropical Pacific is not symmetric. Trade winds blow from east to west along the equator, dragging surface water with them. Warm water piles up in the west — the Indo-Pacific warm pool, the warmest open ocean on Earth — while the east is continually refilled from below by water that has not seen the sun in decades. The result is a basin tilted in two ways at once: sea level stands roughly half a metre higher in the west, and the thermocline, the sharp boundary between the warm surface layer and the cold deep, is far deeper there.

The atmosphere closes the loop. Air rises over the warm pool, where convection is deep and rainfall heavy, travels east aloft, sinks over the cool eastern Pacific, and returns west at the surface as the very trade winds that created the asymmetry. That overturning is the Walker circulation.

The Bjerknes feedback

Jacob Bjerknes noticed in 1969 that this arrangement is self-reinforcing in both directions. Weaken the trades and warm water slides back east; the east-west temperature contrast shrinks; a smaller contrast means a weaker pressure gradient, which weakens the trades further. The same loop run backwards makes a cold event colder.

Writing the coupling crudely, with \(T\) the eastern SST anomaly and \(\tau\) the zonal wind stress anomaly:

$$ \frac{\partial T}{\partial t} \;\sim\; \underbrace{-\,\overline{w}\,\frac{\partial T'}{\partial z}}_{\text{upwelling}} \;+\; \underbrace{-\,u'\frac{\partial \overline{T}}{\partial x}}_{\text{advection}}, \qquad \tau \;\propto\; T $$

Because \(\tau\) grows with \(T\) and \(T\) grows with \(\tau\), the system has positive feedback. Positive feedback alone would run away to one extreme and stay there. It does not — so something must supply a memory that turns the sign around.

El Niño and La Niña

The crucial detail is that upwelling does not stop during El Niño. It keeps running; it simply pulls up water that is already warm, because the thermocline has dropped below the reach of the upwelling. This is why the monitor tracks the depth of the 20 °C isotherm (D20) rather than the strength of the winds alone.

The recharge oscillator, derived

The memory is heat content, and the derivation below is the argument for why. It starts from shallow-water dynamics and ends with a testable number: the lag by which warm water volume leads SST.

1. The ocean the waves live in

Take a 1½-layer reduced-gravity ocean: a warm active layer of mean depth \(H\) over a deep motionless abyss, with reduced gravity \(g' = g\,\Delta\rho/\rho_0\). On an equatorial \(\beta\)-plane, linearised about rest and keeping only long waves:

$$ \frac{\partial u}{\partial t} - \beta y\,v + g'\frac{\partial h}{\partial x} = \frac{\tau^x}{\rho_0 H} - \epsilon u $$

$$ \beta y\,u + g'\frac{\partial h}{\partial y} = 0 $$

$$ \frac{\partial h}{\partial t} + H\left(\frac{\partial u}{\partial x} + \frac{\partial v}{\partial y}\right) = -\epsilon h $$

The second equation is the long-wave approximation: meridional acceleration is dropped, so the north–south flow is in geostrophic balance with the thermocline slope. This is what admits Kelvin and long Rossby waves while filtering out gravity waves, and it is the step that makes the rest tractable.

2. Integrate, and the wind curl appears

Integrate the continuity equation over the equatorial waveguide — across the basin in \(x\), and between \(\pm y_N\) either side of the equator. Zonal divergence integrates to the boundary values, and what survives is the meridional flux through the edges:

$$ \frac{d}{dt}\iint h \,dx\,dy \;=\; -H\!\!\int \big[\,v\,\big]_{-y_N}^{+y_N} dx \;-\; \epsilon \iint h \,dx\,dy $$

Away from the equator that meridional transport is the Sverdrup transport, set by the curl of the wind stress:

$$ V_{\mathrm{Sv}} = \frac{1}{\rho_0 \beta}\,\frac{\partial \tau^x}{\partial y} $$

This is the pivot of the whole theory, and it is easy to read past. The equatorial band gains and loses heat content according to the curl of the wind stress, not its strength. The westerly anomaly of an El Niño peaks near the equator, so it has anticyclonic curl on both flanks, and the resulting Sverdrup transport carries mass poleward, out of the waveguide. Writing \(h_W\) for the western thermocline anomaly that stands in for the basin’s warm water volume, and \(\tau \propto T_E\) for the atmosphere’s near-instant response to eastern SST:

$$ \boxed{\;\frac{dh_W}{dt} = -\,r\,h_W \;-\; c\,T_E\;} $$

The basin discharges while the warm event is still growing. Nothing has to reverse the winds; the discharge is a consequence of the winds that the warming itself produced.

3. The surface layer in the east

A mixed-layer heat budget for the eastern box, keeping the terms that matter there:

$$ \frac{\partial T_E}{\partial t} = \underbrace{-\,\frac{\overline{w}}{H_m}\left(T_E - T_{\mathrm{sub}}\right)}_{\text{upwelling through the mixed layer}} \;\underbrace{-\,u'\frac{\partial \overline{T}}{\partial x}}_{\text{zonal advection}} \;-\; \lambda T_E $$

The subsurface temperature drawn up by that upwelling is not a constant: it depends on how far down the thermocline sits. To first order \(T_{\mathrm{sub}} = \gamma'\,h_E\). This is the thermocline feedback, and it is why the monitor reports D20 at all. During El Niño upwelling does not stop — it keeps running and simply delivers warm water.

4. Eliminating the eastern thermocline

One variable too many. A Kelvin wave crosses the Pacific in about two months and a long Rossby wave returns in six — both fast compared with a cycle of several years — so the basin is close to equilibrium with the wind at any moment. The eastern thermocline is then slaved to the western reservoir plus the local wind forcing:

$$ h_E \;\simeq\; h_W + b'\,\tau \;\propto\; h_W + b\,T_E $$

Substituting, and gathering the growth terms into a single \(a\) — the Bjerknes feedback minus damping, \(a = \gamma b - \lambda\) — leaves a closed pair:

$$ \boxed{\;\frac{dT_E}{dt} = a\,T_E + \gamma\,h_W, \qquad \frac{dh_W}{dt} = -\,c\,T_E - r\,h_W\;} $$

5. Solving it

This is linear, so try \(\mathbf{x} \propto e^{\sigma t}\):

$$ \mathbf{A} = \begin{pmatrix} a & \gamma \\ -c & -r \end{pmatrix}, \qquad \mathrm{tr}\,\mathbf{A} = a - r, \qquad \det \mathbf{A} = \gamma c - a r $$

$$ \sigma = \frac{a-r}{2} \pm \frac{1}{2}\sqrt{(a-r)^2 - 4(\gamma c - ar)} = \frac{a-r}{2} \pm \frac{1}{2}\sqrt{(a+r)^2 - 4\gamma c} $$

That simplification of the discriminant is worth doing by hand once: \((a-r)^2 + 4ar = (a+r)^2\). So when the coupling \(\gamma c\) is strong enough that \((a+r)^2 < 4\gamma c\), the root is complex and the system oscillates:

$$ \sigma = \underbrace{\frac{a-r}{2}}_{\text{growth}} \pm\, i\,\omega, \qquad \omega = \sqrt{\gamma c - \left(\frac{a+r}{2}\right)^{2}} $$

Two things follow. The period \(2\pi/\omega\) comes out at three to five years for observed coefficients. And the growth rate is \((a-r)/2\) — the Bjerknes feedback net of ocean damping. The observed Pacific sits close to neutral, which is why ENSO is irregular and needs weather noise to keep going, rather than ticking like a metronome.

6. The lag, which is the testable part

Now get the phase relationship rather than assuming it. With \(T = \mathrm{Re}[\hat{T}e^{\sigma t}]\) and \(h = \mathrm{Re}[\hat{h}e^{\sigma t}]\), the first equation gives \(\sigma\hat{T} = a\hat{T} + \gamma\hat{h}\), so

$$ \frac{\hat{h}}{\hat{T}} = \frac{\sigma - a}{\gamma} = \frac{1}{\gamma}\left[-\frac{a+r}{2} + i\,\omega\right] $$

The imaginary part is positive and the real part is negative, so the phase angle lies between 90° and 180°: \(h\) leads \(T\). In the weakly damped limit \(a, r \to 0\) it is exactly 90° — a quarter cycle. For a four-year period that is about a year; for the observed three-to-five-year range, nine to fifteen months, or two to three seasons.

This is not a curiosity. It is the entire practical case for watching warm water volume: it is the variable that moves first, and the lag is long enough to be useful. It is also what the phase portrait shows — two quantities a quarter cycle apart trace an ellipse rather than a line, and the sense of rotation says which one leads.

Beware the axes. The loop is anticlockwise with warm water volume on the horizontal and Niño 3.4 on the vertical, which is how the monitor plots it. Transpose them and the identical physics reads clockwise.

What the indices actually measure

The dashboard reports each box as an SST anomaly against the 1991–2020 day-of-year climatology, and a 90-day running mean of Niño 3.4 alongside. That running mean is labelled an ONI proxy and never “the ONI”: the official Oceanic Niño Index is a three-month running mean of monthly ERSSTv5 anomalies on a base period that is refreshed every five years. The two track each other closely and are not the same number, and a figure that called one the other would be wrong in a way nobody could detect downstream.

\(\text{Niño 3} - \text{Niño 4}\) is reported because it distinguishes flavours: a positive value means the anomaly is concentrated in the east (a canonical, coastal El Niño), a negative one means it sits in the central Pacific (a “Modoki” event). The two have different rainfall teleconnections despite similar Niño 3.4 values.

Why this page also tracks cyclones

El Niño is widely said to “suppress” Atlantic hurricanes. That is true on a seasonal average and misleading day to day: it raises the bar for genesis, it does not veto it. 2004 and 1969 both produced major Atlantic hurricanes during warm events. What matters operationally is whether a window opens — a transient stretch where the hostile conditions relax locally.

The storm as a heat engine

Emanuel’s potential intensity comes from treating a mature cyclone as a Carnot cycle. Air takes up entropy from the sea at temperature \(T_s\), ascends in the eyewall, exports it at the outflow temperature \(T_o\), and the work available is dissipated by surface friction. Equating generation to dissipation gives the maximum wind:

$$ V_{\mathrm{pot}}^{2} = \frac{C_k}{C_D}\,\frac{T_s - T_o}{T_o}\,\big(k_0^{*} - k\big) $$

\(k_0^{*}\) is the saturation enthalpy at the sea surface and \(k\) that of the boundary-layer air, so \(k_0^{*} - k\) is the air–sea thermodynamic disequilibrium — the fuel. The ratio \((T_s - T_o)/T_o\) is the Carnot efficiency, which is why a warm ocean under a cold tropopause is favourable and SST alone is not the whole story.

Why shear alone is not the criterion

A developing vortex has to keep its core saturated. Vertical shear tilts it and drives environmental air through the mid-troposphere and into that core. If the imported air is already moist it costs the storm little; if it is dry it must be moistened before it can ascend, and the energy for that comes out of the storm. So the damage done by ventilation is the product of how much air is imported — which scales with the shear \(u_{\mathrm{shear}}\) — and how dry it is.

Dryness is measured as a non-dimensional entropy deficit, the entropy needed to saturate mid-level air divided by the disequilibrium available to supply it:

$$ \chi_m = \frac{s_m^{*} - s_m}{s_{\mathrm{SST}}^{*} - s_b} $$

The numerator is the deficit of the intruding air; the denominator is the same air–sea disequilibrium that appears in the potential intensity. Normalising this way is deliberate: the same absolute dryness matters more over a cool ocean, which has less to give.

Putting it together

The ventilating tendency scales as \(u_{\mathrm{shear}}\,\chi_m\) and the storm’s ability to resist scales as \(V_{\mathrm{pot}}\). Their ratio is the ventilation index:

$$ \mathrm{VI} = \frac{u_{\mathrm{shear}}\;\chi_m}{V_{\mathrm{pot}}} $$

It is dimensionless by construction — \([\mathrm{m\,s^{-1}}] \times [1] / [\mathrm{m\,s^{-1}}]\) — which is what lets one threshold be compared across basins and seasons instead of being retuned for each.

The three factors fail independently, and that is the operational point. Shear can relax while the mid-troposphere stays too dry for it to matter; a moist column can sit under shear that shreds anything that forms; potential intensity can be ample while either of the others forbids genesis. A shear map alone answers one third of the question, and during El Niño it is routinely the third that is not binding.

Windows on the dashboard are opened and closed by a state machine with hysteresis and a dwell time, not by today’s value crossing a line. A single day below the threshold is noise; the question is whether conditions stay favourable long enough for genesis to occur.

Reading the dashboard honestly

• Every anomaly needs its baseline. The same SST can be +0.4 °C or +0.9 °C depending on the reference period, and the difference is not small compared with the thresholds that name an event.
• D20 comes from a model analysis, not a measurement. It inherits model bias in a way altimetry does not.
• The warm water volume anomaly on the dashboard uses a short, moving baseline — the mean of the record accumulated so far. It is for reading the shape of the trajectory. The quotable numbers are PMEL’s.
• ERA5 lags real time by about five days, so the genesis rows carry their own date and are never today’s.

Back to the monitor · the forecast record