Quantifying variance components in ecological models based on expert opinion.
Expert opinion is often relied on to build ecological models when empirical data are absent. Despite widespread use, expert models often ignore uncertainty though it may affect predictions. We assess the importance of such uncertainty for modelling forest restoration. Of an initial 19 experts found using a literature search and peer recommendations, five parameterized models predicting ecological responses to proposed restoration actions for a degraded dry woodland system in Victoria, Australia. We incorporated uncertainty from three sources: disagreement (between-expert uncertainty), self-assessed imprecision (within-expert uncertainty) and modelled system stochasticity. These sources of uncertainty were quantified as variance components in hierarchical models. The between-expert variance component contributed more to overall model uncertainty than both within-expert variance and modelled system stochasticity. The estimate of between-expert variance also had the greatest parameter uncertainty of the three components. Synthesis and applications. We present a method to decompose variance in model predictions. We suggest that modelling strategies relying on a single expert opinion, consensus between experts, or that only incorporate uncertainty due to system stochasticity could produce biased models and over-confident predictions. Management decisions based on biased and over-confident predictions lead to inefficient conservation investments and poor outcomes. Our research highlights the importance of seeking multiple expert opinions to fully characterize uncertainty and make robust decisions.