Self-breeding: a new method to estimate local Lyapunov structures

Anziam Journal, 2000

Iterative methods are used to generate Lyapunov vectors (lvs) and singular vectors (svs). Their roles in describing atmospheric error growth and predictability are studied. lvs are produced by evolving a set of initially random perturbations and using a modified Gram-Schmidt re-orthogonalisation to ensure their orthogonality. The structures of lvs and svs, and finite-time normal modes (ftnms), are com-

2008

1994

It is frequently asserted that in a chaotic system two initially close points will separate at an exponential rate governed by the largest global Lyapunov exponent. Local Lyapunov exponents, however, are more directly relevant to predictability. The difference between the local and global Lyapunov exponents, the large variations of local exponents over an attractor, and the saturation of error growth near the size of the attractor-all result in nonexponential scalings in errors at both short and long prediction times, sometimes even obscuring evidence of exponential growth. Failure to observe exponential error scaling cannot rule out deterministic chaos as an explanation. We demonstrate a simple model that quantitatively predicts observed error scaling from the local Lyapunov exponents, for both short and surprisingly long times. We comment on the relevance to atmospheric predictability as studied in the meteorological literature.

Nonlinear Processes in Geophysics, 2008

ABSTRACT Due to the chaotic nature of atmospheric dynamics, numerical weather prediction systems are sensitive to errors in the initial conditions. To estimate the forecast uncertainty, forecast centres produce ensemble forecasts based on perturbed initial conditions. How to optimally perturb the initial conditions remains an open question and different methods are in use. One is the singular vector (SV) method, adapted by ECMWF, and another is the breeding vector (BV) method (previously used by NCEP). In this study we compare the two methods with a modified version of breeding vectors in a low-order dynamical system (Lorenz-63). We calculate the Empirical Orthogonal Functions (EOF) of the subspace spanned by the breeding vectors to obtain an orthogonal set of initial perturbations for the model. We will also use Normal Mode perturbations. Evaluating the results, we focus on the fastest growth of a perturbation. The results show a large improvement for the BV-EOF perturbations compared to the non-orthogonalised BV. The BV-EOF technique also shows a larger perturbation growth than the SVs of this system, except for short time-scales. The highest growth rate is found for the second BV-EOF for the long-time scale. The differences between orthogonal and non-orthogonal breeding vectors are also investigated using the ECMWF IFS-model. These results confirm the results from the Loernz-63 model regarding the dependency on orthogonalisation.

Tellus A, 2000

We study the evolution of finite perturbations in a meteorological toy model with extended chaos, namely the Lorenz '96 model. The initial perturbations are chosen to be aligned along different dynamic vectors: bred, Lyapunov, and singular vectors. Using a particular vector determines not only the amplification rate of the perturbation but also the spatial structure of the perturbation and its stability under the evolution of the flow. The evolution of perturbations is systematically studied by means of the so-called mean-variance of logarithms diagram that provides in a very compact way the basic information to analyze the spatial structure. We discuss the corresponding advantages of using those different vectors for preparing initial perturbations to be used in ensemble prediction systems, focusing on key properties: dynamic adaptation to the flow, robustness, equivalence between members of the ensemble, etc. Among all the vectors considered here, the so-called characteristic Lyapunov vectors are possibly optimal, in the sense that they are both perfectly adapted to the flow and extremely robust.

Physical Review E, 2010

Bred vectors are a type of finite perturbation used in prediction studies of atmospheric models that exhibit spatially extended chaos. We study the structure, spatial correlations, and the growthrates of logarithmic bred vectors (which are constructed by using a given norm). We find that, after a suitable transformation, logarithmic bred vectors are roughly piecewise copies of the leading Lyapunov vector. This fact allows us to deduce a scaling law for the bred vector growth rate as a function of their amplitude. In addition, we relate growth rates with the spectrum of Lyapunov exponents corresponding to the most expanding directions. We illustrate our results with simulations of the Lorenz '96 model.

Journal of Nonlinear Science, 1992

We develop methods for determining local Lyapunov exponents from observations of a scalar data set. Using average mutual information and the method of false neighbors, we reconstruct a multivariate time series, and then use local polynomial neighborhood-to-neighborhood maps to determine the phase space partial derivatives required to compute Lyapunov exponents. In several examples we demonstrate that the methods allow one tO accurately reproduce results determined when the dynamics is known beforehand. We present a new recursive QR decomposition method for finding the eigenvalues of products of matrices when that product is severely ill conditioned, and we give an argument to show that local Lyapunov exponents are ambiguous up to order 1/L in the number of steps due to the choice of coordinate system. Local Lyapunov exponents are the critical element in determining the practical predictability of a chaotic system, so the results here will be of some general use. Key words, experimental chaotic time series, local Lyapunov exponents, global predictability, local predictability of chaos.

Physical review. E, Statistical physics, plasmas, fluids, and related interdisciplinary topics

We describe methods of estimating the entire Lyapunov spectrum of a spatially extended system from multivariate time-series observations. Provided that the coupling in the system is short range, the Jacobian has a banded structure and can be estimated using spatially localized reconstructions in low embedding dimensions. This circumvents the "curse of dimensionality" that prevents the accurate reconstruction of high-dimensional dynamics from observed time series. The technique is illustrated using coupled map lattices as prototype models for spatiotemporal chaos and is found to work even when the coupling is not strictly local but only exponentially decaying.

2014

The growth of small errors in weather prediction is exponential on average. As an error becomes larger, its growth slows down and then stops with the magnitude of the error saturating at about the average distance between two states chosen randomly. This paper studies the error growth in a low-dimensional atmospheric model before, during and after the initial exponential divergence occurs. We test cubic, quartic and logarithmic hypotheses by ensemble prediction method. Furthermore, the quadratic hypothesis suggested by Lorenz in 1969 is compared with the ensemble prediction method. The study shows that a small error growth is best modeled by the quadratic hypothesis. After the error exceeds about a half of the average value of variables, logarithmic approximation becomes superior. It is also shown that the time length of the exponential growth in the model data is a function of the size of small initial error and the largest Lyapunov exponent. We conclude that the size of the error at the least upper bound (supremum) of time length is equal to 1 and it is invariant to these variables. Predictability, as a time interval, where the model error is growing, is for small initial error, the sum of the least upper bound of time interval of exponential growth and predictability for the size of initial error equal to 1.

2012

The growth of small errors in weather prediction is exponential. As an error becomes larger, the growth rate should diminish. Finally, all systematic growth should stop, and the magnitude of the error should oscillate about a value equal to the magnitude of the distance between two states chosen randomly. The aim of this paper is study of error growth in low-dimensional atmospheric model after the initial exponential divergence died away. For this purpose we test several hypotheses (cubic, quartic, logarithmic) by ensemble prediction method. Also quadratic hypothesis that was introduced by Lorenz in 1969 is compared with the ensemble prediction method. The study shows that a small error growth is best modeled by quadratic hypothesis. After the initial error exceed a half of average value of variables than logarithmic approximation is better.