Abstract

Time series analysis is ubiquitous in paleoclimate research, with applications ranging from identifying periodicities to investigating the temporal continuum to understand how energy within the Earth system is redistributed across various timescales, assigning time, filtering to highlight specific features in climate datasets, detecting regime shift, and identifying coherent spatiotemporal variability between multiple independent time series. These types of analyses require sophisticated expertise to identify appropriate methods for a given dataset and set the parameters, preparing data for analysis, and specifying the null hypothesis. Here, we present autoTS, an automated machine learning system for time series analysis that capture the strategies and methods that experts use and that systematically searched through the space of solutions for a given dataset.

Date

January 1, 1970

Authors

Deborah Khider, Feng Zhu, Yolanda Gil

Journal

AGU fall meeting abstracts

Volume

2019

Pages

PP43D-1637