SmoothyrSmooth air, forecast

Method & accuracy

How the forecast works, and how well

The full method is published here, along with an accuracy record that includes misses.

Accuracy record

After each flight, expert reports and a daily sample of automated outlooks are checked against pilot and aircraft turbulence reports (PIREPs and AIREPs) from the Aviation Weather Center. A report counts as matching when it falls within 50 nautical miles of the route, 4,000 feet of cruise altitude and 90 minutes of the forecast time.

Data

InputUse
ECMWF IFS 0.25° forecast, 00/06/12/18 UTC runsWinds, temperature and height at 400, 300, 250, 200 and 150 hPa (about FL240 to FL450); convective energy (CAPE); surface gusts at both airports. Licensed CC-BY-4.0 by ECMWF.
Aviation Weather Center pilot reportsRecent turbulence reports near the route, and after-flight verification.
OurAirportsAirport locations (public domain).

Method

The automated outlook follows the approach of the Graphical Turbulence Guidance (GTG) system developed at NCAR: compute several clear-air turbulence diagnostics from a weather model, convert each one to the same EDR scale, then combine them (Sharman & Pearson, 2017). It is an independent implementation on ECMWF data. It is not the NCAR GTG software and not a NOAA product.

1. Sample the route

The cruise portion of the great-circle route is sampled about every 140 km, with each point timed at an assumed cruise speed of 830 km/h. At each point, four neighbouring points about 50 km away provide horizontal wind gradients.

2. Compute diagnostics for each layer

For each of the four layers between the pressure levels:

  • Vertical wind shear (VWS)
  • Ellrod index TI1 = VWS × deformation
  • Ellrod index TI2 = VWS × (deformation + convergence)
  • Inverse Richardson number (1/Ri, from static stability and shear)

3. Map to EDR and combine

Each diagnostic D is mapped to EDR with the lognormal method of Sharman & Pearson (2017):

ln EDR = C1 + C2 · (ln D − μ_D) / σ_D

C1 = −2.953 and C2 = 0.602 are the in situ EDR climatology for 20,000–45,000 ft. μ_D and σ_D come from a global sample of ECMWF forecasts. The four EDR estimates are averaged with equal weights. The calibration script is part of the project, so the fit can be rerun and checked.

4. Rate the ride

Categories follow the ECMWF clear-air turbulence thresholds:

RatingEDR (m2/3 s−1)
SmoothBelow 0.10
Light0.10–0.20
Moderate0.20–0.45
Severe0.45 and above

The headline rating is the roughest level reached over at least two consecutive points (about 150 km or more), so a single isolated value does not set the rating for the whole flight.

5. Blend with NOAA GTG over the United States

Over the contiguous United States, for flights within NOAA's 18-hour GTG window, each route point averages the ECMWF engine with NOAA's operational GTG forecast for the same layer. In our first side-by-side check against 1,311 pilot reports, this blend separated moderate-or-greater reports from smooth ones better than either input alone.

How the data are updated

A scheduled job runs every three hours. It collects each new Euro (ECMWF) open-data forecast, issued four times a day, and the latest NOAA GTG run, computes the turbulence fields on a global half-degree grid, and stores them. Your forecast is read from these stored grids, so results are consistent for everyone checking the same flight. If a flight falls outside the stored forecast window, the site computes the same diagnostics directly from ECMWF point data instead.

Known limitations

  • Convective turbulence. Bumps in and near thunderstorms are flagged from model instability, not forecast directly. Crews usually route around storms.
  • Mountain waves. Not modelled in the automated outlook. Expert reports check for them.
  • Vertical resolution. The public pressure-level data are about 1.5 km apart near cruise altitude, which smooths sharp shear layers. Native-level ECMWF data would reduce this.
  • Calibration. The current calibration is provisional. It comes from a short global sample, not a season-long match with aircraft EDR data. Weights will be tuned as verification data build up.
  • Route and timing. Actual routes, altitudes and delays differ from the great-circle assumption.

References