Clear-Air Turbulence

Aircraft weather radar sees water. Clear-air turbulence has none, which is why a 900 km/h collision with a kilometer-thick shear layer arrives without warning. Severe clear-air turbulence (CAT) over the North Atlantic rose 55% between 1979 and 2020, and the mechanism — a warming tropics steepening jet-stream shear — is one of the most directly felt consumer consequences of climate change.

How it works: Kelvin-Helmholtz at altitude

CAT lives at the boundary between two air masses moving at different speeds. Above some shear threshold, the interface ripples into vortices (the Kelvin-Helmholtz instability, the same physics that breaks ocean waves), then cascades into smaller turbulent eddies. A mature CAT patch is roughly 1–3 km thick and 10–100 km long, with no moisture to scatter radar. Aircraft fly into it the way a boat hits a wake it cannot see.

The instability is governed by the Navier-Stokes equations, the unsolved Clay Millennium Prize problem. Statistical prediction works. First-principles prediction of which air parcel turns turbulent next does not.

The climate signal

Prosser, Williams et al. (Geophysical Research Letters, 2023) measured a 55% rise in severe North Atlantic CAT over 41 years of reanalysis data — the strongest signal yet found. The driver: the tropics warms faster aloft than the poles, the equator-pole temperature gradient changes, the jet stream picks up shear. Williams (Reading, 2017) projected that a doubling of atmospheric CO₂ would deliver roughly +60% light CAT, +95% moderate, +150% severe. Current US serious-injury rate is around 50 per year. On Singapore Airlines SQ321 (May 2024) a 73-year-old passenger died over Myanmar after the aircraft dropped 54 meters in seconds; medical reporting blamed cardiac arrest during the encounter.

Detection: what 120 seconds buys

Doppler LIDAR fires UV laser pulses ahead of the aircraft and measures the velocity of air molecules from the backscatter. Range at cruise altitude: about 30 km, which at 900 km/h is 120 seconds of warning. That is enough to seat the cabin crew and turn on the seatbelt sign. It is not enough to deviate the flight path (5–10 minutes minimum). Retrofit cost — tens of thousands per airframe, plus nose-cone real estate currently held by weather radar — has so far exceeded the actuarial cost of injuries.

The proactive layer is machine learning on pilot reports (PIREPs) and accelerometer-derived Eddy Dissipation Rate (EDR, the IATA crowd-sourced system operational from 2020). XGBoost and LightGBM models in 2024–2025 benchmarks beat the FAA's Graphical Turbulence Guidance (GTG) by 18–25% on moderate-or-greater categories. NOAA fine-tuned Google DeepMind's GraphCast in 2024 on its Global Data Assimilation System and improved the upstream wind-shear forecast that feeds GTG.

What's contested

The 55% figure depends on the ERA5 reanalysis model's representation of upper-level shear, because CAT is too sparse for direct satellite observation. Skeptics including Robert Sharman (NCAR) have argued part of the signal reflects improved reanalysis resolution since 2000. The climate-CAT mechanism is settled; the magnitude of the anthropogenic share of the trend is contested at roughly the ±20% level.

The ML models predict turbulence-generating conditions, not turbulence patches. The 30-minute to 6-hour horizon (the operational sweet spot for re-routing) is still the hardest target. No current system reliably forecasts a specific CAT patch at a specific waypoint.

Why this has to do with other realms

The Navier-Stokes equations also govern concept brain turbulence — whole-brain dynamics measured by fMRI transfer entropy that predict antidepressant response with AUC 0.70 (Molecular Psychiatry, 2025), using the same eddy-dissipation vocabulary as atmospheric science. Atmosphere and cortex are both Navier-Stokes systems operating near criticality, both measured by what dissipates, both invisible without instrumentation. In September 2025 Gómez-Serrano and DeepMind (arXiv 2509.14185) found new families of unstable singularities in the Euler equations — possibly the first structural crack in the problem in 180 years. If CAT prediction ever becomes deterministic, it will be downstream of that crack. So will brain turbulence.

An open question

LIDAR gives 120 seconds. Route planning needs 30 minutes. What closes the gap — every airframe networking real-time EDR reports into a moving turbulence map, or an atmospheric model that finally cracks the Kolmogorov cascade from first principles?

Key sources

Prosser, Williams et al., "Evidence for Large Increases in Clear-Air Turbulence over the Past Four Decades" (Geophysical Research Letters, 2023) — the load-bearing trend paper.
Williams, "Increased light, moderate, and severe clear-air turbulence in response to climate change" (Advances in Atmospheric Sciences, 2017) — the projection everyone quotes.
IPCC AR6 WGI, Chapter 3 (2021) — the equator-pole gradient mechanism.
Sharman & Pearson, "Prediction of Energy Dissipation Rates for Aviation Turbulence" (Journal of Applied Meteorology and Climatology, 2017) — the EDR foundation.
arXiv 2509.14185 (Gómez-Serrano et al., DeepMind, September 2025) — Euler singularity families.
To verify: 2025 npj Climate and Atmospheric Science paper on LightGBM for global low-level turbulence.

Abhishek