Increasingly, attempts are being made to predict the responses of natural ecosystems to global changes in climate and land use. I argue that the model of vegetation dynamics underlying much of the current debate is too simplistic to yield predictions that will be useful at a scale relevant to most land-use decisions. I trace the history of ideas on vegetation dynamics from early concepts of succession to climax through to more recent formulations that indicate that vegetation mosaics can consist of patches in alternative stable states, with transitions between states mediated by disturbance, management and climatic events, either singly or in combination. Examples of the possible occurrence of multiple stable states are increasing in the literature, in a range of different vegetation types. State and transition models aim to capture the dynamics of systems with metastable states and event-driven transitions. As yet these models are entirely descriptive and non-quantitative. I compare a state and transition model and a quantitative Markov model, both developed for Mediterranean landscapes in California. The Markov formulation does not allow for event-driven transitions since transition probabilities have to be constant over time, while the state and transition model provides no quantitative estimate of transition probabilities. I propose that a useful modeling approach will combine both approaches, using the state and transition approach to modify transition probabilities in a standard Markov formulation. By linking this approach into a GIS (Geographic Information System) format, spatially-explicit landscape modeling which incorporates spatial and temporal changes in transition probabilities is possible. Such models are required if we are to provide effective predictions of landscape change which will be useful in a management context.
