A calibrated diagnostic of clear air turbulence (CAT) in the free troposphere and stratosphere has been developed in the ECMWF Integrated Forecast System (IFS) using Cy47r1 and the new moist physics for Cy48r1. The diagnostic largely follows the method proposed by Sharman and Pearson (2017), whereby predictors for turbulence are projected onto the climatological distribution of the EDR (m2=3s-1) whis is defined as the cube root of the turbulent eddy dissipation rate.
We computed several indices from daily 0-24 h high resolution (9 km) forecasts of the IFS and compared during January to March 2019 their distribution to civil aircraft data from the NOAA Meteorological Assimilation Data Ingest System (MADIS) in the height range of 5-12 km using model level data. In addition, the IFS ensemble was run at 18 km resolution during the first 14 days of January 2019 allowing to evaluate the ensemble mean and probabilistic skill of turbulence forecasts. It is found that a CAT based on a modified Ellrod1 index and/or the total dissipation rate that is derived from the subgrid physical momentum tendencies of the IFS, provide useful guidance for severe turbulence. Point correlations with observations from the high-resolution system of 0.34 and 0.31 (0.35 combined) were obtained for the two CAT (EDR) products and a mean absolute error of 0.055 m2=3s-1. Encouraging results come from the ensemble forecasts with ensemble mean correlations above 0.4 for January and a continuous rank probability score of below 0.03 m2=3s-1. Overall, the EDR point correlations are in between the point correlations of 0.53 for 10 m wind speed and 0.2-0.4 for daily tropical rainfall over land as obtained from 24h forecasts.
Given archiving and computing constraints and the needs of a direct validation of the IFS turbulence scheme, we decided to only put the CAT based on the total dissipation rate into operations. This product should also be of interest in a future ERA6 climate reanalysis. The information provided in this document is also intended to enable user specific postprocessing including the computation of CAT based on Ellrod1, as well as advanced postprocessing using non-linear regression and/or machine learning.