A Bayesian Belief Network Advisory System for Aspen Regeneration

Abstract

Abstract The probability-based Bayesian Belief Network (BBN) methodology is demonstrated to be an alternative to rule-based methods in forest management expert systems. Unlike a rule-based system, a BBN incorporates uncertainty in the knowledge and input data without sacrificing knowledge modularity. After reviewing the graph and probability theory needed to define a BBN as a joint distribution representable by a directed acyclic graph, an 11-variable network modeling Rocky Mountain aspen sucker density response to different management options is constructed. For a typical aspen site, the model's estimated marginal probabilities of sucker response exhibit values consistent with those expected. Finally, a BBN is shown to be fairly tolerant to small parameter errors, and a new method is given for BBN model validation. For. Sci. 37(2):627-654.