An interactive modelling tool to support knowledge elicitation using extreme case models

Abstract

Knowledge elicitation can be a crucial aspect of modelling. When few data are available, it enables predictions to be made on the basis of expert knowledge. It also provides the opportunity for stakeholders to express their understanding of a system to help assess a model and help ensure that their point of view is accounted for. In this paper, we describe an interactive modelling tool to help express and evaluate stakeholders' knowledge about water requirements of floodplain and wetland vegetation (Figure 1). It aims to maximise the breadth of views to which the user is exposed, and minimise mandatory user input. This helps prompt the user to reflect on their knowledge and empowers them to decide what they feel confident in claiming. This is achieved by automatically generating extreme case models (with different parameter values) for the user to evaluate even before they have given any input. Visualisations of these results prompt the user to provide information that constrains the models. These constraints take the form of key concepts of knowledge about suitability, namely the bounds (e.g. ideally, river red gums require 3-8 months of flooding) and relationship between any two points (e.g. 2 months flooding is better than 1 month flooding). This tool helps to capture uncertainty in elicited knowledge by identifying constraints rather than single models and expecting knowledge to be changeable and evolving. This contrasts with approaches that develop multiple consensus solutions, within which dissenting and novel understandings might be suppressed, and approaches that elicit uncertainty as measurable probabilities or possibilities which are themselves uncertain. Although we use a habitat suitability model as an example, this method is generic and can be used in many other applications eliciting relationships among variables.