Multi‐model ensemble predictions of aviation turbulence

Abstract

Turbulence remains one of the leading causes of aviation incidents. Climate change is predicted to increase the occurrence of clear‐air turbulence and therefore forecasting turbulence will become more important in the future. Currently, the two World Area Forecast Centres (WAFCs) use deterministic numerical weather prediction models to predict clear‐air turbulence operationally; it has been shown that ensemble forecasts improve the forecast skill of traditional meteorological variables. This study applies multi‐model ensemble forecasting to aviation turbulence for the first time. It is shown in a 12‐month global trial from May 2016 to April 2017 that combining two different ensembles yields a similar forecast skill to a single model ensemble and yields an improvement in forecast value at low cost/loss ratios. This finding is consistent with previous work showing that the use of ensembles in turbulence forecasting is beneficial. Using a multi‐model approach is an effective way to improve the forecast skill and provide pilots and flight planners with more information about the forecast confidence, allowing them to make a more informed decision about what action needs to be taken, such as diverting around the turbulence or requiring passengers and flight attendants to fasten their seatbelts. The multi‐model ensemble approach is intended to be made operational by both WAFCs in the near future and this study lays the foundations to make this possible.

Multi‐diagnostic multi‐model ensemble forecasts of aviation turbulence

“…However, as in Storer et al . (), statistical significance with the 95% confidence interval cannot be shown. Both the upper and lower confidence interval bounds for the multi‐model ensemble are shown in Table and there are only six occasions where the multi‐model ensemble is significantly more skilful than MOGREPS‐G, only one occasion for EPS51 and there are no occasions when it is significantly higher than both.…”

“…It is interesting to note that EPS12 seems to be slightly less skilful than EPS51, which is what was found by Storer et al . () but here it is consistent across all turbulence diagnostics.…”

“…As in Storer et al . (), probabilistic forecasts were created, but this time for five turbulence diagnostics (Ellrod and Knapp () Turbulence Index 1, Brown Index, mountain wave turbulence, Richardson number and convective precipitation accumulation) and two ensembles: Met Office Global and Regional Ensemble Prediction System (MOGREPS‐G) and European Centre for Medium‐Range Weather Forecasts (ECMWF) Ensemble Prediction System (EPS). By combining the predictors, a multi‐diagnostic predictor was created.…”

“…These forecasts are derived from the output of deterministic models at a resolution of 1.25° with seven vertical levels (ICAO, ). However, previous studies have shown the improved performance in turbulence forecasting of probabilistic forecasts derived from ensemble models (Gill and Buchanan, ; Storer et al ., ).…”

“…Storer et al . () showed that a single‐diagnostic multi‐model ensemble was able to increase forecast skill and relative economic value, although not at a statistically significant level. However, they also concluded that, by combining the two ensembles, a single authoritative forecast is created with an increased operational resilience.…”