Implementing Bayesian network models using expert elicitation for instream barrier assessment

Le résumé fourni par la source

Abstract Reducing river fragmentation is crucial for restoring freshwater biodiversity. Cost‐effective methods of assessing the likelihood of fish passage at river infrastructure are required for spatial planning of barrier mitigation strategies. A paucity of empirical data on the swimming capabilities and movement behaviour of many fish species presents a challenge for evaluating barrier permeability. We used a combination of expert knowledge and empirical data to define prior probabilities in Bayesian network (BN) models to estimate culvert, ford and weir permeability for multi‐species fish assemblages in New Zealand. The models have been implemented as part of a national fish passage assessment tool. Model outputs are illustrated for a range of structures. Uncertainty associated with incomplete knowledge was explicitly incorporated in the BN models. Experts were most confident in predicting barrier permeability under conditions where fish passage was expected to be poor. Estimates of fish passage success were more varied for conditions considered less likely to impede fish movements. This reflects experts' understanding of the varying swimming and climbing capabilities of different fish species and life stages and how this impacts barrier permeability. The BN models form the basis of a new fish passage assessment tool that has been implemented in New Zealand. Users collect data using a mobile app, recording the key features of instream structures that have been determined to influence the likelihood of successful passage. The BN models are then used to objectively classify the risk to fish passage for each structure. Synthesis and applications. We have demonstrated that BN modelling and expert knowledge can be used for assessing the likelihood of fish passage where empirical data are lacking or sparse. The probabilistic framework is consistent with the need to reflect that many structures are partial barriers to fish movement and cannot be represented accurately in a binary pass/fail classification. This facilitates risk‐based decision‐making, but also suggests that widely used connectivity indices should not ignore the uncertainty in barrier permeability by parameterising connectivity as a single value.

Ce résumé expose les affirmations des auteurs. BNTIC ne l’interprète pas comme une validation indépendante des résultats.