The paper investigates characterizing predictability of the atmosphere. Main part is definition of nonlinear Lyapunov exponent (NLLE) and its saturation property - the Predictability limit of relative growth of initial error.
Then it is compared to Rosenstein method of estimation of maximal Lyapunov exponent. Approximations of these methods are presented.
Preliminary results are shown for both methods computed from time series of daily mean temperature at the 500 hPa level of ERA-40 reanalysis.