The development of reliable modeling techniques to predict freshwater species is important as many become increasingly scarce. A number of modeling techniques have been proposed for predicting populations of freshwater fish and invertebrates based on variables that describe the habitat and surrounding environment. Bayesian Belief Networks (BBNs) are an established modeling technique within the Artificial Intelligence (AI) community and are proposed as an alternative to neural networks and other methods commonly used in this domain. In BBNs variables are represented by nodes. Arcs connect related variables according to their causal links allowing evidence or data to propagate through the network in a fashion determined by the causal structure. BBNs have an advantage over some other modeling methodologies in that the model produced is bi-directional allowing the user to interrogate the model from cause to effect and vice versa. BBNs easily allow for models constructed using a mixture of expert knowledge and data dependent methods, are tolerant in conditions of incomplete data and can be applied to small datasets.

Results/Conclusions

This paper describes how data on New Zealand stream invertebrate populations and habitats in the lower North Island were used to construct a BBN model. The model was tested using an independent data set and the quality of the results were assessed using the area under the receiver operating characteristic curve (AUC).