Using hidden Markov models to inform conservation and management strategies in ecosystems exhibiting alternative stable states

Successful conservation of ecosystems exhibiting alternative stable states requires tools to accurately classify states and quantify state transition risk. Methods that utilize early warning signals are promising approaches for helping managers anticipate impending state transitions, but they require high-resolution temporal or spatial data for individual sites, and they do not directly implicate causes or quantify their impacts on transition risk. There is need for a modelling approach that can assess state transition risk with lower resolution temporal data, identify drivers of state transitions and assess the impact of changes in these drivers on the persistence of ecosystem states. We developed a novel, integrated modelling framework that (a) classifies states in a way that reflects the qualitative dynamics of catastrophic regime shifts, (b) estimates state transition probabilities and (c) uses annual survey data to identify top predictors of state transitions and quantify their effects on transition risk. We applied our model to short time series from 123 shallow lakes that exhibit clear- and turbid-water alternative stable states. We found that clear lakes were more likely to transition to the turbid state as total phosphorus levels increased or occurrence of submerged vegetation decreased. Additionally, increases in planktivorous or benthivorous fish biomass elevated transition risk. Too few turbid-to-clear transitions were observed to identify predictors of transitions in this direction. Synthesis and applications. Our study will inform conservation and management strategies for ecosystems exhibiting alternative stable states by providing a new tool to accurately classify states, compare state transition risk among sites based on resilience and system perturbations, and identify key variables to target to prevent undesirable transitions. Although we focus on shallow lakes as a case study for our modelling approach, we emphasize that our framework is applicable to any ecosystem known to exhibit alternative stable states.