MODELLING FORESTRY EFFECTS ON STREAM ECOSYSTEMS USING A BAYESIAN BELIEF NETWORK.

J.M. Quinn¹ and W.J. Walley². ¹NIWA, PO Box 11115, Hamilton, New Zealand, ²Staffordshire University, The Octogon, Beaconside, Stafford, ST18 0DG, England

Managing impacts of logging activities on stream ecosystems requires an understanding of the complex processes involved. Bayesian Belief Network (BBN) models can deal with this complexity, but are simple enough to be used by resource managers and forest harvest planners. A BBN is an expert system in which the cause-effect relationships between variables are defined and conditional probabilities are set for each variable being in each of its possible states (normally 3 or 4 per variable) given the state of “parent” variables. When evidence is presented to the network, in terms of the actual state of one or more variables, the beliefs in the states of all the other variables are updated by propagating the effect of evidence in a stepwise way throughout the whole network. We developed a BBN model to predict forest harvest effects on stream habitat and invertebrate indicators based on data from a survey of 46 Coromandel Peninsula streams during summer 1999, literature information, and experience gained during eight years of monitoring logging effects. We will demonstrate how this can be used to guide logging planning to manage environmental impacts and to diagnose the likely causes of observed impacts.