EWSmethods: an R package to forecast tipping points at the community
level using early warning signals and machine learning models
Abstract
Early warning signals (EWSs) represent a potentially universal tool for
identifying whether a system is approaching a tipping point, and have
been applied in fields including ecology, epidemiology, economics, and
physics. This potential universality has led to the development of a
suite of computational approaches aimed at improving the reliability of
these methods. Classic methods based on univariate data have a long
history of use, but recent theoretical advances have expanded EWSs to
multivariate datasets, particularly relevant given advancements in
remote sensing. More recently, novel machine learning approaches have
been developed but have not been made accessible in the R environment.
Here, we present EWSmethods – an R package that provides a unified
syntax and interpretation of the most popular and cutting edge EWSs
methods applicable to both univariate and multivariate time series.
EWSmethods provides two primary functions for univariate and
multivariate systems respectively, with two forms of calculation
available for each: classical rolling window time series analysis, and
the more robust expanding window. It also provides an interface to the
Python machine learning model EWSNet which predicts the probability of a
sudden tipping point or a smooth transition, the first of its form
available to R users. This note details the rationale for this
open-source package and delivers an introduction to its functionality
for assessing resilience. We have also provided vignettes and an
external website to act as further tutorials and FAQs.