Bayesian network models for environmental flow decision-making: 1. Latrobe River Australia

Current methods underpinning environmental flow (eFlow) decisions often lack transparency, do not adequately consider uncertainties and rarely include adaptive management principles. We report the development and application of an eFlow Bayesian Network (BN) model that links four flow components with an ecological model to predict the spawning and recruitment of two important native fish species, the Australian Grayling and River Blackfish, in the highly regulated and flow-stressed lower Latrobe River in Victoria, Australia. Autumn high flows, in conjunction with low stream temperature, are critical for Grayling spawning. The BN model was used to predict the probability of spawning and recruitment of these two native fish species for four flow scenarios. Quantitative data, flow simulation models and expert judgement were used to parameterize the BN model. The model results showed clearly that currently, and into the future, there is a very low likelihood of spawning and recruitment of Australian Grayling in the lower Latrobe. River Blackfish are minimally affected by the predicted reductions in flow and increased stream temperatures. Management scenarios aimed at modifying flows and stream temperatures to increase the likelihood of successful spawning and recruitment of Australian Grayling were assessed. Self-sustaining populations of Australian Grayling could conceivably be achieved in the upper reaches of this river if fish passage was provided through an on-stream reservoir. A major benefit in building and applying an eFlow BN model is that it can facilitate meaningful analysis and discussion of the ecological effects of particular eFlow regimes.