The transmission dynamics data set totaled 14,255 juvenile
salmon nonlethally assayed for copepodid, chalimus, and motile stage
lice at 1- to 3-km intervals along 40–80 km of three different
migration routes containing two to three farms each (Fig. 1
and Fig. 4, which is published as supporting information on the PNAS
web site). From among three candidate models, the data best supported a
model that had a uniformly distributed ambient population of infectious
planktonic larvae and point sources of planktonic larvae situated at
the farms (likelihood ratio test, *P* ≪ 0.0001; Akaike weights ≈1; Tables 1–3, which are published as supporting information on the PNAS web site). Across all
data sets, this model fit the data well (Fig. 2
and Figs. 5 and 6, which are published as supporting information on the
PNAS web site). The other models contained only farm- or ambient-origin
lice. With the parameter estimates from the best-fit model, we
reconstructed the spatial distributions of infective larvae originating
from each source. Farm salmon were the primary source of lice, raising
the density of infective parasite larvae above ambient levels for
>80 km of the migration route (Figs. 2, 5, and 6).

Figure 1Study area and sample sites for one of the data sets (April 28 to May 8, Tribune Channel;

Fig. 2).
Approximately 50 pink and 50 chum salmon were collected at each sample
site (stars) and nonlethally assayed for sea lice. The remaining
Tribune Channel data sets had a similar structure. The three active
salmon farms under study are identified by filled squares. An
additional farm (white square near the western end of Tribune Channel)
could have contributed lice but was excluded from the analysis because
of its peripheral position relative to the sample sites. Fallow and
smolt farms are not shown. Gilford Island is situated east of northern
Vancouver Island, BC, Canada.

Figure 2Sea lice transmission dynamics and survival of juvenile chum salmon (

*A*) and pink salmon (

*B*)
migrating past three active salmon farms. The seaward migration of
salmon is from left to right, and the farm locations are shown by
vertical dotted lines in the first row. The data were collected along
the Tribune Channel migration corridor in 2004 (see

Fig. 1).
The three columns correspond to three replicate sets of samples taken
April 18–28 (TR-I), April 28 to May 8 (TR-II), and May 21–29 (TR-III),
2004 (note the change in scale). The first row shows the estimated
spatial distributions of planktonic copepodids originating from all
sources (thick gray line), from farm salmon (three thin curves), from
ambient sources (horizontal thin line), and the second generation of
farm-origin lice (dashed curve, TR-III only). Reproduction of lice
parasitizing the juvenile salmon was not considered in TR-I and -II
because of the absence of gravid female lice in those data sets. The
middle three rows depict the mean abundances of lice (±95% bootstrap
confidence interval) and maximum-likelihood model fits (black lines)
along the migration route for the developmental progression through
parasitic copepodid, chalimus, and motile stages. The bottom row
depicts the estimated remaining juvenile salmon population that
survived sea lice infestation. Temperature and salinity were measured
at each site and averaged 9.0°C and 30.2‰ (TR-I), 10.4°C and 26.1‰
(TR-II), and 12.3°C and 22.2‰ (TR-III).

The
data from the survival experiments totaled 3,687 juvenile salmon with
initial infection intensities ranging from zero to five chalimus lice.
The data best supported a survival model that contained a
gamma-distributed random variable for the parasite's developmental
stage at which there is a marked increase in pathogenicity (likelihood
ratio test; pinks, *P* = 5 × 10^{−18} and chums, *P* = 5 × 10^{−35}; Fig. 3; see also Fig. 7 and Table 4, which are published as supporting information on the PNAS web site). We mapped the survival
model onto space via the average juvenile salmon migration speed (≈1 km·day^{−1};
Table 3) and coupled it to the larval distributions and infection rates
identified by the transmission dynamics model. By removing ambient lice
from the best-fit model, we calculated the proportions of the juvenile
salmon populations that survived parasitism from farm-origin lice.
These were 5–26% for pink salmon and 10–70% for chum salmon in the
Tribune Channel data sets, 49–78% for pink salmon and 69–91% for chum
salmon in the Knight Inlet data sets, and 11–35% for pink salmon in the
Kingcome Inlet data sets. Detailed results are available in *Supporting Text*, which is published as supporting information on the PNAS web site.

Figure 3Survival of juvenile chum salmon over a range of sea lice abundances. Sixty juvenile chum salmon initially infested with

*H*
_{0}
lice (all copepodids or chalimus I/II) were introduced into
flow-through ocean enclosures and provisioned with salmon feed. Each
image corresponds to an individual enclosure. The black line shows the
trajectory for the daily number of survivors. The light-gray lines are
the trajectories of 1,000 simulations of the best-fit model. The model
was simulated as a Markov chain tracking the number of survivors in
time. Each day, the number of mortalities was drawn from the number of
survivors on the previous day using a binomial distribution with
mortality probability calculated from the best-fit survival model. For
all treatment replicates, the model has the same parameter values,
except for

*H*
_{0}, which is specific to each enclosure.