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

The Salmon Parasite in Linnhe, Lorn and Shuna (SPILLS) project focused on testing and improving sea lice dispersal monitoring and modelling techniques. The SPILLS Project developed and compared different models designed to predict distribution of sea lice in Scottish sea lochs.

Synthesis of Key Findings and Recommendations

1. Overall, the sea lice dispersal models evaluated in the Project replicated the variability of observational data and produced a reasonable match to the most suitable infection pressure (sentinel cage) data.

All three sea lice dispersal models' predictions of sea lice infestation pressure were in broad agreement with sentinel caged fish sea lice count data in the Loch Linnhe study area, although there were inevitably some discrepancies. The best fit occurred when the signal-to-noise ratio was relatively high.

Model evaluation using pelagic larval sea lice counts in both study areas was not successful (see item 4 below).

2. Calibration, testing and sensitivity analysis of the individual components of sea lice dispersal models (both hydrodynamic and particle tracking models) is the best approach to ensuring robust model results.

Models and field data should be used together to strengthen monitoring and modelling methodologies and improve understanding of real-world sea lice larvae distributions.

Calibration and validation of hydrodynamic models is of particular importance because of the sensitivity of particle tracking results to hydrodynamic modelling outputs, and the relative difficulty in validating biological elements directly.

Standard tests are available that can be used for assessing the accuracy of particle tracking models in simulating advection and diffusion (e.g. Brickman et al., 2009).

Calibration and testing would benefit from the development of standard tests for the accuracy of particle tracking model simulations of lice behaviour (e.g. swimming behaviour), in addition to a standard protocol for sensitivity analysis

Testing showed that the use of passive, neutrally buoyant particles in particle tracking models does not represent sea lice larvae distributions well. Including sea lice larvae swimming behaviours improved model predictions.

3. Sentinel caged fish sea lice counts currently provide the best data for validation of sea lice dispersal model predictions.

Sentinel caged fish sea lice counts provide a useful time-integrated signal of the infection risk to fish and an indirect inference on distributions of copepodid life stages, which are an output of sea lice dispersal models.

A good fit was found between sea lice dispersal model predictions and sentinel caged fish lice counts for two of the models when lice were relatively abundant (autumn 2011 in the Loch Linnhe study area), such that there was a high signal-to-noise (prediction error) ratio. The fit of the third model was not significant; therefore, this model requires further development. When the signal-to-noise ratios were lower, the model fits to data were less good, as the ability of models to detect signals in the higher relative noise is reduced (Silver, 2012).

Development of a more geographically comprehensive sentinel cage deployment programme may presently be the most appropriate way of providing wider validation of these and similar models, provided that good data are available on lice numbers in salmon farm cages.

4. Observations of pelagic sea lice larvae concentrations obtained using the implemented monitoring techniques are not suitable for use in validating sea lice dispersal models.

Sea lice were detected in 4.8 % out of a total of 372 collected zooplankton samples.

Modelling results predict that sea lice concentrations in the sea are patchy and transient.

The level of variability in observed sea lice larvae concentrations was comparable between model and observational data.

Current sea lice monitoring techniques sample small volumes of water over short time periods. The mismatch between the fine-scale resolution but limited coverage (in time and space) of monitoring and the much coarser resolution of predicted sea lice concentrations produced by sea lice dispersal models means that achieving good agreement between observed sea lice concentrations and model predictions is probably unrealistic.

Capturing planktonic sea lice larvae and identification by microscope is very resource intensive due to the small size of the larvae, and relative low average abundance within high density and diverse zooplankton communities.

The limitations and potential bias of all field data sources should be recognised and understood (Skogen et al. 2022). Failure to capture and definitively quantify sea lice larvae does not equate to an absence in the water column. As such, care must be taken to avoid concluding from a negative result that no larvae were present in the sampling area at the time of sampling.

5. The Project was unable to draw conclusions about the potential for sea lice counts on wild-caught salmonids to help validate sea lice dispersal models.

There were significant logistical difficulties in collecting sea lice counts from wild-caught salmonids due to scarcity of the fish and a lack of detailed knowledge of their movement within Shuna Sound. Consequently, the Project was unable to use the data collected to help in validating the models.

Further work is needed to assess whether it is practical to design and operate a suitable wild fish monitoring programme (as has been done in Norway) and utilise the data obtained to help validate sea lice dispersal models.

6. An ensemble modelling approach can provide a means of improving understanding of areas of least, and greatest, uncertainty in model predictions. This can improve confidence in decision-making (Oidtman et al. 2021).

An ensemble approach enables identification of geographic areas of agreement and disagreement between predictions given by different particle tracking models (or different plausible configurations of a single particle tracking model). The approach can help identify where model predictions are most and least coherent.

The use of an ensemble approach is more resource-intensive than using a single, well-calibrated and validated particle tracking model. More research development is required to assess the practicality of using an ensemble approach of multiple simulations for management decisions in Scotland.

7. Targeted laboratory and field experimentation to assess sea lice behaviours, such as diurnal vertical migration, could help improve calibration and accuracy of particle tracking models.

Vertical distributions of sea lice larvae are fundamental to dispersal patterns and must therefore be correctly reproduced by models. This could be assessed using laboratory experiments or field observations (e.g. Nelson et al. 2018).

Development of improved methods for capturing and identifying sea lice larvae from the plankton would facilitate field studies.

Novel methods showing early promise for improving identification of sea lice in samples include automated image analysis, fluorescence microscopy, and eDNA, ideally in conjunction with large-volume depth stratified sampling.

8. Models validated at one location may have to be used in locations for which observations are lacking. Where this is the case, model predictions should be treated with appropriate caution.

Model predictions of sea lice concentrations should be validated using suitable field data wherever possible.

Good data for validation may not always be feasible to obtain. There are currently limits to the suitability of planktonic field data (e.g. due to patchiness) for use in validation, and the collection of extensive field data can be resource intensive.

Where validated models must be applied in another location without equivalent field validation, appropriate calibration and sensitivity analysis of the individual model components remains essential.



Back to top