Modeling dynamics of adult female lice at salmon farming sites in Eastern Canada: a stochastic, state-based approach

Introduction: Sea lice are parasitic copepods that harm salmon health, reduce farm productivity, and create ecological and economic challenges for aquaculture. Methods: A stochastic, state-based, time-dependent epidemiological model was developed to characterize the dynamics of adult female sea lice (Lepeophtheirus salmonis) infestation in Atlantic salmon farms in New Brunswick, Canada. The model integrated covariates associated with farming practices and environmental conditions (stocking week, farming cycle week as proxy of fish age, sea lice treatments, seaway distance to neighboring farms as a proxy for waterborne transmission, and sea surface temperature). Data from 57 farming sites were used for model training and validation. An initial exploratory analysis assessed the relationship between treatment timing and recovery from infestation. Treatment effects were incorporated into weekly transitions between infestation states, accounting for severity and time-varying environmental factors. Results: Results suggest that spring and summer stocking increases exposure to external infestation pressure and raises the probability of high lice concentrations. Further, reduced winter treatments are associated with elevated infestation levels. Treatment effectiveness appeared to be compromised by continued waterborne transmission from nearby farms. Discussion: The model achieved an overall likelihood of 59%, reaching up to 74% during the first 10 weeks following stocking. Limitations included the use of proxy connectivity measures, i.e. seaway distance, rather than hydrodynamic connectivity, and the absence of data on fish size, salinity, and other farming practices such as fish density. Additionally, we were unable to include information from all farms in the study area, potentially underestimating transmission risk. Addressing these gaps and integrating hydrodynamic connectivity and fish growth models could improve predictive performance.