Salmon Parasite Interactions in Linnhe, Lorn and Shuna (SPILLS): final project report

Foreword

Sustainable use and management of the marine environment requires robust environmental monitoring and modelling to understand and mitigate against risks caused by human activities, including potential risks posed by sea lice originating from salmon farms in Scotland to wild and farmed salmonids.

Marine finfish farms are regulated by the Scottish Environmental Protection Agency. Modelling and environmental monitoring underpin SEPA's approach to regulating discharges from fish farms (SEPA 2019) and will also underpin its regulation of the interactions between sea lice from farms and wild salmonids as it takes on this responsibility in due course.^[1]

The SPILLS project compared three models that had hitherto been developed by different organisations as tools to predict levels and distribution of sea lice in the environment as they spread from open-pen salmon aquaculture facilities. Such models have a role to play in understanding transmission of lice between fish farm sites and also to wild fish. This is the first attempt at a detailed configuration, calibration and validation exercise to compare the models. It has used best available information on actual lice levels in the environment and also set out to explore potential for collecting further such data as a source for fine-scale validation.

In addition to analysing model predictions from individual models, the project also applied an averaging "ensemble" approach as part of a research exercise to explore potential for combining model outputs. Use of an ensemble, together with geographic information tools, enabled an approach for visualising how coherence in model predictions varies throughout a sea loch i.e. where, locally, do the models tend to make similar or divergent predictions.

Key findings of interest to the development of sea lice models for regulation are:

1. Overall, the models worked reasonably well when evaluated against time-integrated (sentinel caged fish) data with a sufficiently strong signal (high signal-to-noise ratio). Therefore, targeted sampling for model validation should be focused on time periods when relatively high lice numbers are expected.

2. Planktonic sea lice data, collected using plankton trawls and pumps, did not produce useful data with which to evaluate model performance because lice distributions are so patchy in space and time.

3. There was variation in predicted local lice densities among the three models tested. Hence, for management applications, either model selection will be required, or a number of models can be averaged in an "ensemble".

4. Calibration and testing of the individual model components is essential; in particular, thorough calibration and validation of the underlying hydrodynamic model is of importance because of the sensitivity of particle tracking results to hydrodynamic modelling outputs, and the relative difficulty in validating biological elements directly.

5. Model predictions are improved relative to simulations of passive particles when sea lice biology is represented; targeted laboratory experiments and further field studies to improve understanding of sea lice can be expected to help improve model parameterisation.

6. Model evaluation in the project was hampered by uncertainties in numbers of lice on farms and the deployment periods of sentinel cages. Had better data been available, it is likely that model fits would have been improved by reducing "noise". A new programme of sentinel cage deployments to provide "clean" datasets would help with the wider validation of these models.

7. Overall, the project has made very substantial advances by bringing together three key modelling groups (research, government and industry) for a detailed exploration of their available models and clear signposting of next steps and cautions required for management applications.