Forest Inventory Modelling using Ontologies and Bayesian Networks

Abstract

As Bayesian networks are statistical models that are easy to interpret and in which domain knowledge can be included explicitly, they are well suited for environmental sciences as much domain knowledge is available and interpretability can lead to interesting new insights. However, Bayesian networks can take much effort to construct. This workload can be reduced by drawing their structure from ontologies. This research explores whether ontologies can help a small forestry company to model their data in a Bayesian network. It does so by merging two existing ontologies to create an ontology. This ontology is transformed into the graph of a Bayesian network. An overview of the structure of methods making this transformation is given. Apart from combining existing methods, steps are added to this structure by drawing from the experiences in creating a Bayesian network in this research. Lastly, further steps in the development of these methods for small companies are identified.