‘mixglm': an R package for estimation of stable states, tipping points, and ecosystem resilience using mixture models

A number of modelling frameworks exist to estimate resilience from ecological datasets. A subset of these frameworks seeks to estimate the whole ‘stability landscape', which can be used to calculate resilience and identify stable states and tipping points. These methods provide opportunities for insights into possible causes and consequences of variation in ecosystem resilience and dynamics. However, because such models can be complex to implement, there has so far been a substantial barrier to their application in ecological research. Here, we present the ‘mixglm' package for R software, which parametrizes stability landscapes using a mixture model approach. It provides tools for the calculation of resilience, identification of stable states and tipping points, as well as visualization functions. Flexible model specification allows the mean, precision, and probability of each mixture component to be linked to multiple predictors, such as environmental covariates. ‘mixglm' is based on Bayesian inference via NIMBLE and supports normal, beta, gamma, and negative binomial distributed response variables. We illustrate the use of ‘mixglm' with a published case of tree cover in South America, which reports a stability landscape with distinct stable states. Using ‘mixglm', we replicated the identification of these states. Moreover, we quantified the uncertainty of our estimates, and computed resilience estimates of South America's forests. We also conducted a power analysis to provide guidance regarding required sample sizes. ‘mixglm' can be readily used to describe stability landscapes and identify stable states in most spatial datasets, and it is accompanied by tools for the calculation of resilience estimates.