Towards robust predictions of salmon lice infestation pressure in Scottish Lochs: Developing a multi-biophysical model ensemble

To assess lice loads and potential impacts on wild fish, hydrodynamic particle tracking models have been developed globally, though their accuracy varies due to differences in structure, input data, and assumptions—particularly when modelling larval stages. In Scotland, multiple biophysical models are regularly used, each demonstrating strengths and weaknesses depending on application. Validation against field data, including plankton tows and sentinel cage results, revealed inconsistencies in individual model performance. Given these discrepancies and the limitations of the observational dataset, an ensemble approach was used to characterise model uncertainty. The ensemble produced more stable predictions and reduced the influence of extreme model outputs, although improvements in predictive accuracy could not be conclusively demonstrated. Further, the ensemble provides a practical way to characterise uncertainty, although improved accuracy relative to individual models cannot be demonstrated with the available data. This ensemble approach enhances risk management in aquaculture and supports informed decision-making, especially when paired with transparent communication of model uncertainty—both from data limitations (aleatoric) and model design choices (epistemic).