335 – SPEED: Reliable Statistical Learning and Data Science
Robust Mode-Detection Schemes in a Non-Stationary Environment
Francois A. Marshall
Department of Mathematics and Statistics of Queen’s University
David J. Thomson
Department of Mathematics and Statistics of Queen’s University
This paper reviews the methods of established mode-detection techniques in the field of multitaper spectrum estimation. The paper presents results of the application of these techniques to the time series of a relative ionospheric opacity meter, an instrument used to measure fluctuations in the opacity of the lower ionosphere to radio waves. The statistical test is robust because of two reasons. First, extraneous measurements in the dataset are replaced by hard-rejection techniques. Second, the test statistic is a spectrum estimator of the component process with the least signal-to-noise ratio, and so its distribution is variance-efficient with respect to a mixture of chi-squared distributions. In the presented analysis, the mode-detection test reveals that a fraction of the noise-like component process of the time series is explicable by periodic phenomena of known origin. In particular, some of the variance of this process is explained by the coupling of solar modes to the radio-emissions opacity of the lower ionosphere.