Abstrakt
Predictability curves show average growth rate of initial and model error of meteorological quantities predicted by numerical weather models. A curve displays limits of predictability and average error of prediction for a chosen day.
The curves are calculated to show effects to predictability of different parameterizations, different resolutions, numbers of ensemble members, initial conditions etc. Low-dimensional atmospheric models are used to carry out predictability studies that would be too expensive to perform using numerical weather prediction models.
This article tests the ability of the Lorenz's (2005) chaotic model to simulate predictability curve of the ECMWF model by quadratic hypothesis. Similar predictability curves are found for the Lorenz's model with N = 120 variables and the ECMWF model from 1990s, for the Lorenz's model with numbers of variables between N = 120 and N = 240 and the ECMWF model from around 2000 and for the Lorenz's model with N = 240 variables and the ECMWF model from around 2010.
Usability and challenges of quadratic hypothesis are also discussed.