STRUCTURE AND CAUSES OF VEGETATION CHANGE IN STATE AND TRANSITION MODEL APPLICATIONS

State and transition (ST) descriptions of rangeland vegetation dynamic s provide information on current perceptions of explicit causes of cha nge in dominant vegetation. Structural attributes of ST applications a llow an evaluation of the complexity of the ST model and comparisons w ith the organization of the traditional succession-retrogression model of secondary succession. An analysis of 29 applications of the ST mod el revealed consistent trends. The number of transitions connecting st ates showed a less-than-expected increase with the size of the applica tion. This is probably associated with limitations to interpret comple x relationships and a need to produce relatively simple applications. Larger applications exhibited a shift towards stable states with pivot al positions within structures less connected (i.e., with fewer transi tions) than expected by chance for a given number of states. Thus, som e stable states assume hey intermediary roles as the number of states considered increases. It is debatable whether this is a property of la rger systems or an effect of modeling bias. The analysis of causes of vegetation change confirmed current perceptions about the importance o f man-related sources of disturbance. Grazing, fire, and control of wo ody plant species are visualized as the most relevant man-related agen ts of change. Some ST applications retain autogenic behaviors embedded in transitions in spite of the event-driven nature of the approach. H owever, the ST model removes autogenic processes from their central ro le as general causes for vegetation change. This approach is theoretic ally very limited because no general properties or attributes of the c omponents (e.g., plant species assemblages, individual species) or pro cesses (e.g., growth, reproduction, mineralization) of the system are used in any comprehensive way to generate predictive rules of wider th an local relevance. Alternative approaches are suggested that would al low ecological generalizations and comparisons across systems.