Detecting and Classifying Events in Noisy Time Series

Time series are characterized by a myriad of different shapes and structures. A number of events that appear in atmospheric time series result from as yet unidentified physical mechanisms. This is particularly the case for stable boundary layers, where the usual statistical turbulence approaches do not work well and increasing evidence relates the bulk of their dynamics to generally unknown individual events. This study explores the possibility of extracting and classifying events from time series without previous knowledge of their generating mechanisms. The goal is to group large numbers of events in a useful way that will open a pathway for the detailed study of their characteristics, and help to gain understanding of events with previously unknown origin. A two-step method is developed that extracts events from background fluctuations and groups dynamically similar events into clusters. The method is tested on artificial time series with different levels of complexity and on atmospheric turbulence time series. The results indicate that the method successfully recognizes and classifies various events of unknown origin and even distinguishes different physical characteristics based only on a single-variable time series. The method is simple and highly flexible, and it does not assume any knowledge about the shape geometries, amplitudes, or underlying physical mechanisms. Therefore, with proper modifications, it can be applied to time series from a wider range of research areas.