Embracing uncertainty: Ensemble models of sea lice larval dispersal

Sea lice, particularly Lepeophtheirus salmonis, pose significant ecological and economic challenges to salmon aquaculture. Dispersal occurs via planktonic larval stages which may spread widely, driven by interacting physical and biological processes. Direct observations are challenging, and biophysical particle tracking models are commonly used to estimate infestation pressure. Nevertheless, there is substantial uncertainty in model parameterization. We assessed ensemble approaches that aim to harness this uncertainty to improve predictions of farm-associated sea lice distributions. Using public data from salmon farms in Scotland (2021–2024) and 20 parameterizations, we compared three ensemble methods: a simple average, a model with spatially varying blending weights, and a machine learning model. Ensemble performance was assessed against that of the constituent parameterizations, all using adult-equivalent infestation pressure (rolling sum of copepodid pressure adjusted for on-fish demographics) to predict the reported mean adult female lice per fish. Cross-validation showed superiority of machine learning ensembles for dynamics on farms, though more direct data on larval concentrations would be required to predict larval infestation pressure. On average, the ensemble model with spatially varying weights outperformed all constituents, particularly for spatial patterns, and predicts ensemble infestation pressure as a latent variable. The simple average ensemble performed better than the median constituent. A resampling experiment confirmed the robustness of these results across constituents and ensemble sizes (3–15). Ensemble modelling is thus a promising pathway toward improving predictions of sea lice infestation pressure, with flexibility to tailor ensembles toward particular goals. Better predictions of sea lice dispersal enable better management of infestations, ultimately improving the welfare and production of farmed fish and reducing lice burdens for wild salmonids.