Abstract |
The seminal paper of Lorenz (1969) established that a fluid system has
limited predictability if the spectral distribution of its energy decays
sufficiently slowly as the scale of motion decreases (i.e., if the
kinetic energy per unit wavenumber has a -5/3 spectrum). According to
Lorenz, predictability is limited by the upscale cascade of errors
beginning at the smallest scales that are not subject to viscous
dissipation. Thus, the flapping of a butterfly's wings might be
supposed to change the weather on a relatively short time scale.
Nevertheless, numerical weather prediction has been reasonably
successful without accounting for the behavior of butterflies. We will
review the various modifications to the notion of atmospheric
predictability that have been proposed since Lorenz's paper and present
new results from recent ensemble simulations of strong winter storms.
We will show that after a few modifications, Lorenz's model is able to
do a surprisingly good job of describing the loss of predictability in
the ensemble forecasts. In the process, we will see why butterflies
are of no practical importance. |