How Do Tornadoes Form?
The spin does not start vertical. It starts lying on its side, rolled up by wind that changes with height, and something has to stand it upright.
Three steps: roll, tilt, squeeze
Junior level — plain language, no maths
Start with something that sounds unrelated. If the wind near the ground blows slowly and the wind higher up blows faster, the air in between gets rolled - exactly the way a pencil rolls if you push the top of it faster than the bottom. You end up with an invisible tube of spinning air lying on its side, like a rolling pin floating above the field.
Now a thunderstorm arrives. A thunderstorm is a powerful updraught, air rushing upwards fast enough to hold hailstones in the sky. When that updraught reaches the rolling tube, it lifts one part of it - and the tube tilts, until what was lying on its side is now standing upright, spinning around a vertical axis. The storm now has a rotating core, several kilometres wide, turning slowly.
Slowly is the important word, because a few kilometres of slow rotation is not a tornado. The last step is the one that turns it into one: the rotating air gets stretched upward and therefore squeezed inward. Anything spinning that gets pulled inward has to spin faster - the same reason a skater speeds up by pulling their arms in, and it is not an analogy, it is the same conservation law. Squeeze a two kilometre rotation down to a hundred metres and the wind speed goes up enormously.
One more thing has to be true. If the cloud base is very high, the funnel spins up but never reaches the ground - you get a rope hanging in the sky and nothing else. Move the sliders and find the combination where all three steps work and the base is low enough for it to touch down.
Things worth knowing
- A spinning skater pulling their arms in is doing exactly what a tornado does. Conservation of angular momentum is not a metaphor here - it is the same equation, applied to air instead of arms.
- A high cloud base is one reason many rotating storms never produce a tornado. The rotation is there, spinning happily, and it simply never reaches the ground.
- Only about one rotating supercell in five produces a tornado. The rotation is common; getting it stretched down to the surface is not.
Vorticity tilting, the mesocyclone, and the role of the cold pool
Student level — the core equations
The quantity being tracked is vorticity, the local spin of the fluid. Vertical wind shear generates horizontal vorticity: with a wind profile \(u(z)\), the horizontal vorticity is \(\eta = -\partial u/\partial z\), typically around 0.01 s⁻¹ for a strong shear profile. That horizontal vorticity is not a tornado and cannot become one until it is reoriented. The tilting term in the vorticity equation does exactly that: an updraught with a horizontal gradient of vertical velocity converts horizontal vorticity into vertical vorticity.
The result is a mesocyclone, typically 2 to 10 km across, rotating at perhaps 20 m/s - impressive on radar and nowhere near a tornado. The amplification comes from the stretching term. Conservation of angular momentum in a converging column means \(\omega r^2\) is approximately constant, so contracting the radius by a factor of ten multiplies the rotation rate by a hundred. A mesocyclone at 3 km radius squeezed to 300 m does exactly that.
Where the near-surface rotation comes from is more subtle than the classic picture suggests. Tilting by the main updraught happens well above the ground, and rotation aloft does not by itself produce a tornado. Current understanding gives a central role to the rear-flank downdraught and its cold pool: baroclinic generation of horizontal vorticity along the boundary of the cold outflow, which is then tilted downward and brought to the surface before being stretched by the updraught. The temperature deficit of that outflow matters - too cold and it undercuts the storm, killing it.
The practical forecasting parameters follow from this. Storm-relative helicity in the lowest kilometre measures the streamwise vorticity available for tilting; CAPE measures the updraught strength available for stretching; the lifting condensation level, which tracks low-level moisture, tells you whether the outflow will be cold enough to undercut. The composite indices used operationally are essentially products of these three.
Key Formulas
| Horizontal vorticity | \(\eta = -\dfrac{\partial u}{\partial z}\) | made by shear |
|---|---|---|
| Tilting term | \(\dfrac{D\zeta}{Dt} \supset \eta\dfrac{\partial w}{\partial x}\) | the updraught stands it up |
| Stretching | \(\dfrac{D\zeta}{Dt} \supset \zeta\dfrac{\partial w}{\partial z}\) | |
| Angular momentum | \(\omega_1 r_1^2 = \omega_2 r_2^2\) | |
| Updraught speed | \(w_{\max} \approx \sqrt{2\,\text{CAPE}}\) | halved in practice by entrainment |
Things worth knowing
- Contracting a rotating column by a factor of ten in radius multiplies its rotation rate by a hundred, because angular momentum goes as r². That single factor is the difference between a mesocyclone and a tornado.
- The rear-flank downdraught has to be cold enough to generate vorticity but not so cold it undercuts the updraught. Tornado formation sits in a narrow window of outflow temperature.
- Doppler radar sees the mesocyclone, not the tornado - the tornado itself is usually below the radar beam at any distance. Warnings are issued on the rotation above, which is why false-alarm rates are high.
Streamwise vorticity, the tornadogenesis problem, and why warnings still fail
Scholar level — full mathematical depth
01Streamwise versus crosswise vorticity
Not all horizontal vorticity is equally useful. Decompose it relative to the storm-relative flow: the streamwise component, parallel to the inflow, is tilted into vorticity that is positively correlated with the updraught, producing a rotating updraught. The crosswise component tilts into a couplet straddling the updraught, which merely splits the storm. This is why storm-relative helicity, which integrates streamwise vorticity along the inflow, discriminates supercell environments far better than bulk shear magnitude. It also explains the strong preference for right-moving storms in the northern hemisphere: the curved hodograph that supplies streamwise vorticity favours one member of the split pair.
02The near-ground problem
The central difficulty is that tilting by the updraught necessarily occurs above the surface, and a vortex aloft does not make a tornado. Something must generate or transport vorticity into the lowest tens of metres. Two mechanisms compete in the literature: baroclinic generation along the cold pool boundary followed by downward tilting in the rear-flank downdraught, and, in the more recent high-resolution simulations, vorticity produced within the surface friction layer itself and swept into the corner flow. Idealised simulations that omit surface drag frequently fail to produce a tornado at all, which is a strong hint that friction is not a detail here but part of the mechanism.
03The corner flow and the swirl ratio
The structure of the vortex near the ground is governed by the swirl ratio, the ratio of tangential to radial inflow. At low swirl the vortex is a single narrow core with the strongest winds just above the surface, in a corner-flow region where the boundary layer separates. Above a critical swirl ratio the core breaks down and the vortex becomes a two-celled structure with downdraught at the centre, which then becomes unstable to multiple subvortices - the discrete damage swaths seen in the strongest events. The most extreme winds ever measured, around 135 m/s by mobile radar, occurred in such subvortices rather than in the parent circulation.
04Why the warning problem is genuinely hard
Radar resolves the mesocyclone but usually not the tornado, so warnings are issued on a precursor that is neither necessary nor sufficient. Roughly 20% of mesocyclones produce tornadoes, and the false-alarm ratio for warnings has hovered around 70% for years. Improvement is now being sought from dual-polarisation signatures such as the tornadic debris signature, which confirms a tornado is on the ground but offers no lead time, and from phased-array radar with much shorter update cycles. The physics constrains what is achievable: if the near-ground vorticity source is partly frictional and partly baroclinic, both are below the beam and neither is observable at range.
05Rating by damage, and what that hides
The Enhanced Fujita scale rates by damage indicators, not by measured wind, because direct wind measurements are almost never available. This creates a systematic bias: a violent tornado that crosses open farmland cannot be rated above EF1, since there is nothing present whose failure would indicate more. Climate trend analyses are affected accordingly - the apparent decline in violent tornadoes is partly a change in rating practice, and the clearest robust signal is not in annual counts but in the clustering of events into fewer, bigger outbreak days.
Key Formulas
| Storm-relative helicity | \(\text{SRH} = \int_0^{h}(\mathbf{v}-\mathbf{c})\cdot\boldsymbol{\omega}\,dz\) | |
|---|---|---|
| Vorticity equation | \(\dfrac{D\boldsymbol{\omega}}{Dt} = (\boldsymbol{\omega}\cdot\nabla)\mathbf{v} + \nabla\times\mathbf{B} + \nu\nabla^2\boldsymbol{\omega}\) | |
| Swirl ratio | \(S = \dfrac{\Gamma}{2Qa}\) | sets single or two-celled core |
| Baroclinic generation | \(\dfrac{D\boldsymbol{\omega}}{Dt} \supset \dfrac{\nabla\rho\times\nabla p}{\rho^2}\) | |
Things worth knowing
- A violent tornado crossing open farmland cannot be rated above EF1, because the scale rates damage and there is nothing there to damage. Climate trends in tornado intensity are partly an artefact of what happened to be in the way.
- Simulations that leave out surface friction often fail to produce a tornado at all. Drag appears to be part of the mechanism, not a correction to it.
- The fastest winds ever measured on Earth, about 135 m/s, were in a subvortex inside a tornado rather than in the tornado itself.