A formalism for characterizing vegetation responses using classification trees and Dempster-Shafer theory of evidence

Predictive models of vegetation change are urgently needed for range management and conservation purposes. Traditional approaches based an climax equilibrium theory do not fit observed dynamics of important types of grasslands and shrublands. Most alternative community- and species-based models require sophisticated information, are only locally applicable or produce predictions that are not directly interpretable for management decisions. Information currently gathered in management settings is mostly ignored despite its potential utility for predicting species responses to disturbance regimes/events and management practices. This study merges elements of comparative and functional ecology in a conceptual approach to employ management-level knowledge and autecological information in the characterization of vegetation responses. A formalism is proposed for practical applications that includes specific methods for identifying response groups of species and criteria for establishing agreement among expert sources. Classification trees are used for arranging individual species into response groups according to autecological similarities for a set of adaptive attributes and specific responses obtained from expert sources. Profiles of attributes associated with responses to individual disturbance factors are then combined and most probable trends for response groups are generated using Dempster-Shafer mathematical theory of evidence: Predictive rules and attached belief intervals can be produced for every non-empty combination of autecological profiles previously determined for individual disturbance factors. This provides generality and portability across physiognomically similar vegetation types. The species profile approach also has important implications for the conservation of functional diversity and risk evaluation of invasion potential by non-native species. A general application of the formalism to the vegetation of the Edwards Plateau (Texas, USA), involving predictions for the effects of grazing and fire, is implemented and limitations and future developments required are discussed.