Abstract

Even though theoretical considerations and several observations point to the possibility that volcanic tremor may be a low-dimensional deterministic phenomenon, there has been no effort in the past to rigorously establish the presence of nonlinear dynamics. In this study we use volcanic tremor time series recorded at Sangay volcano, Ecuador, in order to find such evidence. Initially, the surrogate data test is employed and nonlinear prediction errors are used as a suitable quantity for such a test. Results indicate that tremor time series yielded slightly higher prediction errors than their corresponding surrogates, implying that they can be modeled better by a linear stochastic process at a 95% confidence level. However, the application of Casdagli’s test to the same data shows reduced prediction errors in smaller length scales that is compatible with low-dimensional chaos. The apparent conflict of the two tests can be resolved by taking a closer look at the prediction errors as a function of neighborhood size in phase space. This shows a sharp drop in the number of points available for nonlinear prediction at small length scales, probably due to the undersampling of the process that generates the tremor signal. This puts constraints on the efforts to analyze the nonlinear dynamics of volcanic tremor and calls for a reevaluation of seismic data acquisition procedures at active volcanoes.