Feature Identification in Timeseries Datasets

We present a computationally inexpensive, flexible feature identification method which uses a comparison of timeseries to identify a rank ordered set of features. Features are identified as simultaneous local maxima of absolute deviation in each timeseries. The analyst tunes the method using their knowledge of the physical context. The method is applied to both a dataset from a moored array of instruments deployed in Monterey Bay, California, and a dataset from sensors placed within a submerged cavern network in Tulum, Quintana Roo, Mexico. The results show that the method allows automated identification of both features which were previously identified by analysts in an ad hoc manner as well as features in unstudied datasets which are worthy of further study.