How consistently do ensemble prediction systems represent the growth of atmospheric uncertainty?

Abstract Recent studies have used ensemble‐spread‐based diagnostics to understand atmospheric sources of uncertainty, and its growth and propagation. We aim to provide confidence in the diagnostics‐based approach, and understand its limitations, by comparing Eulerian and Lagrangian geopotential height uncertainty growth diagnostics calculated from 12 different ensemble prediction systems available in The International Grand Global Ensemble. In addition to a Lagrangian growth‐rate diagnostic that uses the ensemble mean velocity (LGR), an alternative form (LGR2) is derived that includes all the uncertainty transport terms and that, for variables with tracer‐like characteristics, also represents the non‐conservative sources of uncertainty. Good agreement between ensemble prediction systems is found in the magnitude and distribution of all growth‐rate diagnostics applied in the midlatitude and polar regions for a lead‐time range starting at 48 hr and ending at between 96 hr and 192 hr, with the best agreement for the largest ensemble size of 50 members. In these situations, ensemble‐spread‐based diagnostics provide a consistent measure of uncertainty growth. However, at shorter and longer lead times, and in tropical regions, ensemble prediction system dependence is greater. In addition, LGR and LGR2 are compared for the fields of geopotential height and potential vorticity (a tracer‐like variable) using data from the European Centre for Medium‐Range Weather Forecasts operational archive. For geopotential height, LGR2 is found to include a significant non‐advective component; therefore, restricting the use of this diagnostic to variables that have tracer‐like characteristics is recommended. This study provides confidence in, and constraints on, the use of spread‐based diagnostics for understanding the sources and transport of uncertainty growth that limit predictability.