The abundance of the parasitic salmon louse has increased with the growth in aquaculture of salmonids in open net pens. This represents a threat to wild salmonid populations as well as a key limiting factor for salmon farming. The Norwegian ‘traffic light’ management system for salmon farming aims to increase aquaculture production while securing sustainable wild salmonid populations. However, this system is at present solely focusing on mortality in wild Atlantic salmon, while responses of sea trout with different ecological characteristics are not included.

We analyze lice counts on sea trout from surveillance data and use Bayesian statistical models to relate the observed lice infestations to the environmental lice infestation pressure, salinity, and current speed. These models can be used in risk assessment to predict when and where lice numbers surpass threshold levels for expected serious health effects in wild sea trout.

We find that in production areas with the highest density of salmon farms (West coast), more than 50 % of the sea trout experienced lice infestations above levels of expected serious health effects.

We also observed high lice infestations on sea trout in areas with salinities below louse tolerance levels, indicating that the fish had been infested elsewhere but were returning to low-saline waters to avoid lice or delouse. This behavioural response may over time disrupt anadromy in sea trout.

The observed infestations on sea trout can be explained by the hydrodynamic lice dispersal model, which provides continuous estimates of lice exposure along the whole Norwegian coast. These estimates, which are used in Atlantic salmon research and management, can also be used for sea trout.

Synthesis and policy implications: Wild sea trout, spending its entire feeding migration in fjords and coastal areas, is at higher risk than Atlantic salmon to lice infestations from aquaculture. The observed high levels of lice infestation on sea trout question the environmental sustainability of the current aquaculture industry in areas with intensive farming. We discuss the complex responses of sea trout to salmon lice and how the ‘traffic light’ management system may include data on this species.

More than 1100 locations are approved for aquaculture production along the Norwegian coast, but 600-700 are simultaneously active in production. These locations are distributed in 13 management/production areas (hereafter production areas) (Fig. 1). The production areas were defined to minimize cross-dispersion (Ådlandsvik 2015).

We analysed a data set of n=2937 sea trout < 200 g, sampled at 40 different sites in 2019 (the most recent data at the onset of this work) (Fig. 1). The fish were caught in traps and gillnets (17 – 21 mm mesh size) in week numbers 20-31 (mid-May to end of July) with a gradual delay from south to north (Table 1). The sampling thus targeted post-smolts recently migrated out from the rivers, a migration that is delayed from south to north by about 6 – 8 weeks (Kristoffersen et al. 2018; Johnsen et al. 2020).

Table 1. Week, production area and the number of sea trout we counted lice on.

                                              Production area

Week   1   2   3   4   5   6   7   8   9   10  11  12  13

  20   32   0   0   0   0   0   0   0   0   0   0   0   0

  21   31 165 162  73  38   0   0   0   0   0   0   0   0

  22   39  92  45 197 100   0   0   0   0   0   0   0   0

  23    0  65  92 185 259  25  43   0   0   0   0   0   0

  24    0   0   0  43 165  28 121  36   8   0   0   0   0

  25    0   0   0   0  11  26  79  58  39   0   0   0   0

  26    0   0   0   0   0   0   0  42  51   0   0   0   0

  27    0   0   0   0   0   0   0   0  17  57  19   0   0

  28    0   0   0   0   0   0   0   0   0 194   0  32   0

  29    0   0   0   0   0   0   0   0   0  30  15  41  68

  30    0   0   0   0   0   0   0   0   0   0   0  16  40

  31    0   0   0   0   0   0   0   0   0   0   0   0  58

Lice counts and expected health effects on fish

Lice counts on sea trout were performed in the field immediately after collection. Fish from traps were anesthetized before sampling (Benzocaine 200 mg/ml diluted by 15-20 ml/100 l water) and released to the sea after recovery. Trout from gillnets were killed. Lice counts were performed with the fish submerged in a white plastic tub (5-10 l) using a strong headlamp (>500 lumen). Counts were performed by certified personnel and the following categories were quantified: copepodite, chalimus 1, chalimus 2, pre-adult, adult male and adult female. Fish length in mm and mass in gram were recorded. Based on previous studies, we defined infestations of 0 – 0.1 salmon lice per gram as a low dose of salmon lice on sea trout. Doses of 0.1 – 0.3 and > 0.3 were defined as moderate and critical doses, respectively, expected to result in health effects on the fish. Doses above 0.3 lice per gram trigger physiological stress responses with return to freshwater for sea trout < 150 g (Taranger et al. 2015).

Hydrodynamic model for environmental variables

The Lice infestation pressure in the environment was estimated by combining: lice counts from all active aquaculture sites along the Norwegian coast (weekly counts of adult female lice), temperature in 3m depth, monthly number of fish per farm, and a hydrodynamic dispersion model system (Albretsen et al. 2011; Myksvoll et al. 2018; Sandvik et al. 2020).

From the hydrodynamic model, we extracted median values of Lice infestation pressure, Salinity and Current in the upper 2 m of the water column from a 20 km radius around the catch site of each sea trout. We averaged the data over weeks to compare with time periods the fish samples were grouped into (c.f. Table 1), and used Lice infestation pressure, Salinity and Current as explanatory variables in the ZAG models.

Zero Altered Gamma (ZAG) models to predict salmon lice on the fish

We modelled Lice Infestation (lice g^-1) on the fish as a response to variation in the environmental variables Lice Infestation pressure, Current and Salinity. The variable Lice Infestation Pressure was first square-root transformed, subsequently all variables were standardized to zero mean and unit standard deviation before inclusion into our main model:

Lice infestation ~ Lice Infestation Pressure + Current + Salinity, using Site = random     (1)

We ran separate analyses for i) the total number of lice and ii) the number of sessile young stages (copepodids, chalimus I and II only).

The model was implemented in the INLA package (Lindgren & Rue 2015) for R (R-Developmental-Core-Team 2019). Due to the zero-inflated and right-skewed nature of the response variable, the number of salmon lice per gram fish, we used a Zero Altered Gamma (ZAG) random effects modelling framework:

Y_i ~ ZAGμ_i, π_i, or Lice on fish_i ~ ZAGGamma_i, Bernoulli_i                                    (2)

MeanY_i= π_i× μ_i  and  varY_i= π_i × r+ π_i- π_i²× rr × μ_i²                                           (3) 

logμ_i= β₁ × Lice Inf Pressure+ β₂ × Current+β₃ × Salinity+u_i           (4)

logitπ_i=γ₁ × Lice Inf Pressure+ γ₂ × Current+γ₃ × Salinity+v_i          (5)

where Y_i is the observed number of lice on sea trout and follows a ZAG distribution. There are two components in the model: a binary Bernoulli part for lice presence or absence, with the mean π_i and a logistic link, and a gamma part for positive values of lice, with the mean μ_i and a log-link (Zuur & Ieno 2018).

In addition to other model validations, we also performed simulation tests to explore the ability of the ZAG model to cope with the proportion of zeros in the data set (23 % and 30 % zeros for the total and sessile young lice, respectively) and compared the observed versus expected values from the final ZAG model. The model was able to cope with the zeros (21 % and 28 % zeros were the modes in the simulation output, for the total and sessile young lice, respectively, Appendix Fig. I), but showed a compressed set of expected values (up to about 1.5 lice g^-1) compared to the observed values, where a small number of fish had extremely high level of infestations (up to 9.1 lice g^-1) (Appendix Fig. II). We ran a series of cross-validation tests (with a simpler mixed model – lmer in R) to explore i) whether single or few Sites was able to drive the main trends, and ii) whether repeated random sampling of 80 % (‘training set’) of the data could predict the last 20 % of the data (‘test-set’), using the R-package groupdata2. The cross-validation confirmed our main results from the ZAG model (Appendix Figs. IX-XI).