Search

State transitions are changes in ecosystem structure and self-reinforcing feedbacks that are initiated when an exogenous driver variable crosses a threshold. Reversing state transitions is difficult and costly. While some state transitions are relatively rapid, many take years to decades. Outside of theoretical models, very little is known about slower state transitions and how they unfold in time and space. We quantified changes in spatial variance as a mesic grassland ecosystem shifts to a shrub-dominated state, using long-term experiments and simulations that maintain grasslands with annual fires or initiate a state transition to shrub dominance by decreasing fire frequency. In the experiments, the susceptibility to state transitions varied substantially in space. In the less frequent fire treatment, some plots became shrub-dominated around year 20 and grass extirpations began in year 25, but a third of the plots were still grass-dominated in year 37. Variable rates of state transition resulted in increasing spatial variance of grass cover over time, whereas shrub cover variance decreased. In the annually burned treatment, grasses remained dominant and the spatial variance of grass cover declined. In a separate experiment, less frequent fires were maintained for 23 years and then switched to annual fires. The switch to annual fires occurred shortly after grass variance started to increase and a majority of these plots quickly returned to a grass dominated state. In simulations, spatial variance remained low and average grass cover was high under frequent fires. If fire frequency decreased below a threshold, the ecosystem transitioned to shrubland, with a transient increase in the spatial variance of grass cover during the transition between states. Synthesis. Spatial variability in the rate and susceptibility to state transitions is indicative of a system with a patchy spatial structure, high spatial heterogeneity and low connectivity between patches. Increases in spatial variance can serve as an indication that some patches have begun a state transition and that management interventions are needed to avoid widespread transitions. This is one of the first empirical examples where altering management after an increase in spatial variance prevented state transitions.