A4 Refereed article in a conference publication
Probabilistic Multivariate Early Warning Signals
Authors: Laitinen Ville, Lahti Leo
Editors: Ion Petre, Andrei Păun
Conference name: International Conference on Computational Methods in Systems Biology
Publishing place: Cham
Publication year: 2022
Journal: Lecture Notes in Computer Science
Book title : Computational Methods in Systems Biology 20th International Conference, CMSB 2022, Bucharest, Romania, September 14–16, 2022, Proceedings
Series title: Lecture Notes in Computer Science
Volume: 13447
First page : 259
Last page: 274
ISBN: 978-3-031-15033-3
eISBN: 978-3-031-15034-0
ISSN: 0302-9743
eISSN: 1611-3349
DOI: https://doi.org/10.1007/978-3-031-15034-0_13(external)
Web address : https://link.springer.com/chapter/10.1007/978-3-031-15034-0_13(external)
Preprint address: http://arxiv.org/pdf/2205.07576(external)
A broad range of natural and social systems from human microbiome to financial markets can go through critical transitions, where the system suddenly collapses to another stable configuration. Anticipating such transition early and accurately can facilitate controlled system manipulation and mitigation of undesired outcomes. Generic data-driven indicators, such as autocorrelation and variance, have been shown to increase in the vicinity of an approaching tipping point, and statistical early warning signals have been reported across a range of systems. In practice, obtaining reliable predictions has proven to challenging, as the available methods deal with simplified one-dimensional representations of complex systems, and rely on the availability of large amounts of data. Here, we demonstrate that a probabilistic data aggregation strategy can provide new ways to improve early warning detection by more efficiently utilizing the available information from multivariate time series. In particular, we consider a probabilistic variant of a vector autoregression model as a novel early warning indicator and argue that it has certain advantages in model regularization, treatment of uncertainties, and parameter interpretation. We evaluate the performance against alternatives in a simulation benchmark and show improved sensitivity in warning signal detection in a common ecological model encompassing multiple interacting species.