Dealing with high uncertainty in qualitative network models using Boolean analysis

摘要: Models for predicting ecological behaviour typically require large volumes of data parameterisation, which is a problem because are scarce. Qualitative modelling (QM) provides an alternative by exploring the entire range possible parameter values. When value completely unknown, QM invokes Principle Indifference (PoI), example sampling from uniform prior distribution. However, if PoI invoked in this probabilistic way, there may be multiple methods defining space and values it, and, worryingly, two different but equally defensible can lead to predictions about ecosystem responses. We investigated how give rise problems QM, developed method based on Boolean that does not suffer same limitations. used case study involved responses species suppression pest. The unknown model parameters were interaction strengths between species. For standard method, we drew randomly (PoI) other distributions. our new instead simply specified ranges “possible” values, analysis technique summarise predictions. As expected, invoking yielded (species-response probabilities) parameterisation schemes. Sometimes differences enough impact decision-making. In contrast, approach classifies outcomes (species responses) as certain, possible, or impossible. Encouragingly, some consistently resolved under shown fact governed simple rules. Our also identify key whose determine whole-system outcomes. non-probabilistic representation uncertainty circumvents philosophical implementations QM. summarises results way interpretable potentially useful conservation decision makers. A priority future research increase efficiency allow it deal with higher complexity (more interactions).