6533b822fe1ef96bd127cbc9

RESEARCH PRODUCT

FILTERING CHAOS: A TECHNIQUE TO ESTIMATE DYNAMICAL AND OBSERVATIONAL NOISE IN NONLINEAR SYSTEMS

subject

description

Nonlinear dynamical models are frequently used to approximate and predict observed physical, biological and economic systems. Such models will be subject to errors both in the model dynamics, and the observations of the underlying system. In order to improve models, it is necessary to understand the causes of error growth. A complication with chaotic models is that small errors may be amplified by the model dynamics. This paper proposes a technique for estimating levels of both dynamical and observational noise, based on the model drift. The method is demonstrated for a number of models, for cases with both stochastic and nonstochastic dynamical errors. The effect of smoothing or treating the observations is also considered. It is shown that use of variational smoothing techniques in the presence of dynamical model errors can lead to potentially deceptive patterns of error growth.