Estimating the Lyapunov-exponent spectrum from short time series of low precision

We propose a new method to compute Lyapunov exponents from limited experimental data. The method is tested on a variety of known model systems, and it is found that the algorithm can be used to obtain a reasonable Lyapunov-exponent spectrum from only 5000 data points with a precision of ${10}^{\mathrm{-}1}$ or ${10}^{\mathrm{-}2}$ in three- or four-dimensional phase space, or 10 000 data points in five-dimensional phase space. We also apply our algorithm to the daily-averaged data of surface temperature observed at two locations in the United States to quantitatively evaluate atmospheric predictability.