Background

This report details the application of the stream salmonid simulator (S3) to coho salmon (Oncorhynchus kisutch) in the Klamath River, northern California (fig.1). The S3 is a deterministic life-stage-structured population model that simulates daily growth, movement, and survival of juvenile salmonids. Here, we document the application of S3 to coho salmon, focusing on several themes relevant to management, including disease dynamics and variation in life history strategies. We build from previous applications of the S3 model to Chinook salmon (Oncorhynchus tshawytscha) in the Klamath River Basin (Perry, Jones, and others, 2018; Perry and others, 2019), expanding the domain to which the S3 model has been applied.

The S3 model was developed to aid fisheries and water managers in understanding the impacts of alternative management actions on anadromous fish populations (Plumb and others, 2019). The model links the effects of river flow to habitat availability and capacity, which drives density-dependent population dynamics in a series of linked habitat units. The environmental template of each habitat unit is defined by a time series of daily discharges, water temperatures, and usable habitat areas or carrying capacities. Survival, growth, and movement processes are simulated for juvenile salmonids in each of these habitat units.

Coho salmon within the Klamath River Basin are classified in the southern Oregon and northern California Evolutionarily Significant Unit and recognized as threatened under the Endangered Species Act (Federal Register, 1997). The decline of coho salmon in the Klamath River Basin is due to a myriad of factors, including construction of main stem dams and agricultural demand for water. The main-stem Klamath River has extensive hydropower development, including dams constructed during the first half of the 20th century that not only limit fish passage within the basin, but alter river discharge and water temperature. Downstream from the lowermost dam, Iron Gate Dam, decreased summer discharge, elevated water temperatures, and decreased habitat availability and water quality are of particular concern for coho and Chinook salmon.

Coho salmon in the Klamath River Basin are affected by alterations to flow and temperature regimes. Increased water temperatures have been cited as a leading factor limiting the recovery of salmonids in the basin (Bartholow, 2005). Salmonids, including coho salmon, are often observed using thermal refuges provided by cold water tributaries in the Klamath River (Sutton and others, 2007). This behavioral strategy may help to limit the increased metabolic costs associated with elevated temperatures in the main-stem Klamath River. Lower flows can also directly impact coho salmon by limiting the amount of rearing habitat and slowing rates of downstream migration, potentially decreasing survival due to increased exposure to predation and disease (Čada and others, 1997). Alterations to flow regimes not only occur in the main-stem Klamath River but in the major tributaries as well (National Research Council, 2004).

The Scott and Shasta Rivers are two of the primary sources of naturally produced coho salmon in the upper anadromous Klamath River (Chesney and Knechtle, 2015; Knechtle and Chesney, 2016). Flows in these tributaries are impacted by surface-water diversions and groundwater pumping for agriculture (National Research Council, 2004). Loss of riparian buffer, water diversions, and decreased cold water spring inputs have contributed to increases in water temperatures (Nichols and others, 2014). Altered hydrology and elevated temperatures are two primary factors threatening coho salmon produced in the Scott and Shasta Rivers.

Coho salmon can have a complex life history with migratory events often occurring between large main-stem rivers and non-natal tributaries (Lestelle, 2007). In the upper anadromous Klamath River, natural production occurs in both large rivers (for example, Scott and Shasta Rivers) and smaller perennial tributaries. Much has been learned about the life history dynamics of naturally produced coho salmon from the Scott and Shasta Rivers owing to annual juvenile trapping efforts and tagging studies using passive integrated transponder (PIT) tags (Manhard and others, 2018a). Juveniles emigrate from natal tributaries at various life stages, either as age-0 fry and parr in the spring and fall, or as age-1 smolts the following spring. Age-0 migrants that enter the main-stem Klamath River in spring typically seek refuge from rising summer water temperatures in the main stem by migrating into non-natal tributaries (Sutton and others, 2007; Sutton and Soto, 2012). Juvenile coho salmon that use non-natal tributaries will migrate back into the main-stem Klamath River when environmental conditions are more favorable, typically during the following winter or spring periods (Manhard and others, 2018a). After reentering the main-stem Klamath River from non-natal tributaries, these juveniles actively migrate downstream as age-1 smolts. Less is known about the fall age-0 migrants that leave the Scott and Shasta Rivers because screw traps on natal tributaries are not operated during this period. Juvenile coho salmon are often observed entering non-natal tributaries during the winter, and these fish may represent some fraction of fall age-0 migrants; however, the source of these fish is uncertain (Witmore, 2014; Manhard and others, 2018a). Fish that overwinter in the Scott and Shasta Rivers leave during the following spring and actively migrate downstream in the main-stem Klamath River as age-1 smolts.

Naturally produced juvenile coho salmon co-occur with both naturally and hatchery produced Chinook salmon in the Klamath River Basin. Juvenile Chinook salmon occur in high densities as a result of hatchery programs that can release up to 5 million smolts annually, and extant populations having not decreased to abundances as low as coho salmon in the area. Interspecific competition can occur in salmonids, where juveniles are territorial and compete for space or shared resources, which may ultimately result in decreased size or survival (Hearn, 1987; Fausch, 1988). Both coho and Chinook salmon occupy similar habitats when rearing in streams and show similar levels of aggression (Lister and Genoe, 1970; Stein and others, 1972). High densities of juvenile coho and Chinook salmon using similar habitat would suggest that some level of competition may occur in the Klamath River.

Modeling Disease Impacts on Juvenile Coho Salmon

Disease can be a major source of mortality in juvenile Pacific salmon (Oncorhychus spp.), including coho salmon in the Klamath River (Fujiwara and others, 2011; Ray and others, 2014). Mitigating the effects of the myxozoan parasite Ceratonova shasta on coho and Chinook salmon is a particular management concern. The parasite has a complicated life cycle with two distinct spore stages and two obligate hosts, a salmonid and a freshwater annelid (Manayunkia occidentalis). Myxospores released from infected salmonid hosts infect the annelid hosts, and actinospores released from infected annelid hosts infect salmonids. C. shasta infects multiple salmonid species, and genetically distinct strains (genotypes I, II, and 0) vary in specificity and virulence in their respective fish hosts. Although all genotypes appear to be able to infect most salmonid hosts, they differ in specificity and virulence. Genotype 0 is most commonly detected in rainbow trout (Oncorhynchus mykiss) but is not typically associated with mortality (Atkinson and Bartholomew, 2010). Genotype I also is a specialist that infects and causes mortality in Chinook salmon. Genotype II is a generalist and is associated with mortality in coho salmon, sockeye salmon (Oncorhynchus nerka), and chum salmon (Oncorhynchus keta) (Hallett and others, 2012; Hurst and Bartholomew, 2012). Although the parasite infects and causes disease in adult and juvenile salmonids in the Klamath River, disease is associated with mortality primarily in juvenile salmonids. C. shasta induced mortality ranging from 0 to greater than 90 percent of age-0 fish is observed in coho and Chinook salmon in sentinel exposure trials (Hallett and others, 2012; Ray and others, 2012), and evidence indicates that C. shasta can influence population dynamics (Fujiwara and others, 2011).

Management actions have recently been implemented at Klamath River dams to mitigate the effects of C. shasta on salmon populations. These management actions consist of flushing flows to scour the annelid host populations that release actinospores and dilution flows intended to reduce the concentration of actinospores (Alexander and others, 2016; Shea and others, 2016). Although modeling studies have simulated a decrease in population-level mortality associated with these management actions (Plumb and others, 2019; Som and others, 2019), the degree to which these management actions are having population-level impacts is uncertain. Thus, improving our understanding of the population-level response to management actions and the impact of future environmental change requires development of analytical tools tailored for coho salmon in the Klamath River Basin. The S3 model detailed in this report is one such tool that integrates our understanding of coho salmon life history, disease ecology, and variation in biological and environmental conditions to assess the population level responses.

Purpose and Scope

Given the threats to coho salmon in the Klamath River Basin and the importance of assessing alternative management actions, we developed a version of the S3 model specific to coho salmon. This model characterizes the dynamics of juvenile coho salmon in freshwater in relation to environmental and biological conditions. Specifically, this version of the S3 model accommodates alternative life-history strategies of naturally produced coho salmon in the Klamath River Basin that includes overwintering and non-natal tributary rearing. We then use the model to assess disease dynamics and quantify the population-level impacts of C. shasta over the historical period during which spore concentrations have been measured. We also assess the impacts of juvenile Chinook salmon on coho salmon using a density-dependent movement submodel within S3.

Disease Modeling

The temporal dynamics of mortality from C. shasta in salmonids typically have three characteristic traits, described by Ray and others (2014) as delayed onset of mortality following initial exposure, a period of rapid mortality, and a plateau where no additional mortality occurs. To parameterize a coho salmon-specific disease model in S3, we used the same analytical approach described in Perry and others (2019), where we fit a survival cure model to sentinel trial data. We extend the work of Ray and others (2014) by analyzing a more recent set of sentinel experiments that incorporated a range of exposure durations into the study design. Because coho salmon mortality is induced by the genotype II of C. shasta (Atkinson and Bartholomew, 2010), all analysis and simulations are based on measurements of genotype II spores. Here, we briefly describe the sentinel experiments that were used to develop the survival cure model. The sentinel trials were designed to quantify how C. shasta induced mortality is influenced by exposure duration, spore concentration during exposure, water temperature during exposure, and water temperature post exposure. We used data from n = 24 sentinel trials conducted in 2014 and 2015; each trial was conducted by:

We used a survival cure model, similar to the approach used by (Ray and others, 2014) to model survival of fish exposed to C. shasta in experimental trials (table 1). The survival cure model is composed of two components: (1) a “cure” probability, which is the proportion of the population expected to survive the disease, and (2) a time-to-death function for individuals expected to die from disease:

The data required to fit the model is the vector x_i = (t_i, c_i), where t_i is either the recorded time of death of the ith individual or the censoring time (that is, the time when the trial ended if the ith fish remained alive for the duration of the trial), and c_i is the censoring indicator (c_i = 0 if t_i = time of death, c_i = 1 if the t_i = trial duration). The cure model allows both the probability of infection and eventual death, π, and the time-to-death, S(t|death), to be modeled separately as functions of covariates.

For π, the covariates and model structure were determined by a previous analysis of this dataset by Som and others (2019). They found that the best-fit model structure on the total mortality in each sentinel trial included genotype II spore concentration, water temperature during exposure period, water temperature during the post-exposure holding period (holding temperature), exposure duration, and an interaction between exposure temperature and holding temperature. This model structure for π was held fixed across all models evaluating time to death “S(t|death),” but the coefficients associated with the covariates for π were estimated jointly with the covariates for S(t|death).

Given the model structure for π, we used a three-step model selection approach, using Akaike’s information criterion (AIC) to arbitrate between models for S(t|death). First, we fit a model set with alternative distributions for S(t|death) and compared these using AIC. We considered five distributions, including the Weibull, log-normal, gamma, log-logistic, and generalized F distributions. This set of models was fit using all main effects and all possible two-way interactions. Second, we used the most highly supported distribution and fit four models evaluating support for either spore concentration or the natural logarithm of spore concentration. With spore concentration and the logarithm of spore concentration, we fit two models each containing either full covariates with interactions or main effects only. We then chose the model with the lowest AIC and proceeded with our last model selection step. We evaluated interaction terms by removing each term one at a time and keeping only interaction terms that reduced AIC by more than 10 units. Within this set of models, we selected the model with the lowest AIC to evaluate and incorporate into S3.

In addition to providing the parameter estimates for the top model, we present the results from this model in two ways. First, we plot the Kaplan-Meier survival curves for each trial (observed response of fish exposed to C. shasta) and compare these with model predictions. Second, to explore the covariate relationships with survival, we plot predicted survival across a range of values for each covariate.

We used program R (R Core Team, 2019) to fit all models in a maximum likelihood framework using the gfcure package (Zhang and Peng, 2007). We incorporate the estimated effects of physical and biological covariates on C. shasta mortality from the mixture cure model into S3 using a disease submodel (Perry and others, 2019).

S3. Model Structure

As noted above, the S3 model of coho salmon in the Klamath River Basin extends previous S3 modeling efforts focused on Chinook salmon (Perry, Plumb, and others, 2018; Perry and others, 2019). We take advantage of the existing S3 model, including submodels describing the movement, growth, and survival of juvenile salmonids. Information specific to coho salmon is incorporated in numerous ways including disease modeling, natal and non-natal tributary dynamics, and parameterized growth and movement submodels.

The Scott and Shasta Rivers are two coho salmon production tributaries to the Klamath River (Chesney and Knechtle, 2015; Knechtle and Chesney, 2016) for which considerable monitoring and tagging data are available. Therefore, for this version of S3, we focus on modeling the production of juveniles from the Scott and Shasta Rivers, recognizing that other source populations exist (for example, Bogus Creek, Trinity River) and could be incorporated into future modeling efforts.

To develop this application of the S3 model, certain assumptions were necessary given the existing information available for coho salmon in the Klamath River Basin. We rely extensively on previous reports, in particular Manhard and others (2018a), and previous S3 reports (Perry, Plumb, and others, 2018; Perry and others, 2019) for model inputs, structure and parameterization. For instance, to parameterize the growth model, we assume a value for the proportion of maximum consumption in the bioenergetics model (see section, “Growth Submodel,” for details) that implies that food availability does not limit growth. Given that we are unaware of any empirical estimates of the proportion of maximum consumption for coho salmon in the Klamath River Basin, we must rely on such assumptions to construct the model and run simulations. We have tried to highlight areas where assumptions were made, given the minimal or entire lack of available information, so that future work may target these uncertainties and stimulate further parameter estimation and S3 model development.

Non-Natal Tributary Submodel

Coho salmon have complex life-history dynamics in the Klamath River where they can emigrate from natal tributaries at different ages (age-0 or age-1), life stages (fry, parr, smolt), and seasons (spring, fall). Age-0 juveniles that leave natal tributaries in the spring disperse downstream in the main-stem Klamath River and immigrate into non-natal tributaries during the summer. Once they enter non-natal tributaries, they may subsequently reenter the main-stem Klamath River during the winter or over-winter in tributaries until the following spring when they reenter the Klamath River and out-migrate as age-1 smolts. We only consider the non-natal tributary dynamics of spring age-0 fish during the summer period in this report and do not model these dynamics for fall age-0 fish. This decision was made because of the added complexity of modeling simultaneous daily entry and exit of fall age-0 fish that may occur during the winter and spring periods based on limited data and given the assumptions associated with fall age-0 fish in general.

To include these dynamics in the S3 model, we first developed models for the probability of fish entering a non-natal tributary during the summer period. We included the location of 12 non-natal tributaries in the S3 model (table 3; fig. 1). Next, for spring age-0 fish that enter non-natal tributaries, we modeled survival from the median entry time for each tributary until the end of the winter emigration or spring emigration period. Next, we modeled the emigration timing for each non-natal tributary for the surviving fish and determine their size-at-date. Combined, these elements allowed us to simulate the abundance, timing, and size of fish reentering the main-stem Klamath River of those that used non-natal tributaries. As opposed to the daily timestep of S3 dynamics during occupancy of the main stem river, survival in non-natal tributaries is modeled at the monthly timescale and emigration timing is modeled at a weekly timescale with fish surviving to reenter the main stem during either winter or spring emigration periods. To parameterize these dynamics, we relied heavily on the analyses of Manhard and others (2018a), who analyzed juvenile coho salmon PIT tag data collected by various State, Federal, and Tribal agencies.

Table 3.    

Non-natal tributaries considered in S3 coho salmon (Oncorhynchus kisutch) simulations.

[River kilometer refers to where the confluence of the tributary is on main-stem Klamath River. Habitat units are identifiers used internally in S3 to describe discrete habitats]

Table 3.    Non-natal tributaries considered in S3 coho salmon (Oncorhynchus kisutch) simulations.

Tributary name	River kilometer	Habitat unit
Horse Creek	239.16	1394
Tom Martin Creek	231.83	1444
O’Neil Creek	223.19	1497
Fort Goff Creek	206.02	1591
Thompson Creek	199.67	1643
Independence Creek	152.9	1894
Ti Creek	130.36	2027
Sandy Bar Creek	123.6	2062
Stanshaw Creek	122.47	2067
Irving Creek	120.7	2072
Whitmore Creek	100.44	2177
Waukell Creek	5.15	2611

Manhard and others (2018a) used conditional binomial models to estimate factors affecting the probability of fish immigrating into non-natal tributaries from the Klamath River. We used the approach detailed above in the S3 Model Structure section to convert from the conditional binomial probabilities used in Manhard and others (2018a) to unconditional multinomial probabilities. We used the second lowest AIC model from Manhard and others (2018a) that included maximum weekly decrease in discharge and the mean weekly temperature of the Klamath River at the mouth of each tributary as covariates (see Manhard and others, 2018a, for parameter estimates). This model describes factors that influence the summer entry into non-natal tributaries. To match the same scale as Manhard and others (2018a), we used the model to predict weekly immigration probabilities, then spread these probabilities evenly across the week to match the daily timestep of S3 dynamics during occupancy in the main stem river. The daily number of fish moving past tributary mouths are then multiplied by these daily non-natal immigration probabilities to simulate the number of fish entering each tributary from each upstream habitat unit. These fish are accumulated in each non-natal tributary as the simulation progresses, removing them from the main-stem Klamath River dynamics. Fish that do not enter non-natal tributaries continue with the full S3 simulation dynamics.

Next, we modeled winter and spring, where fish that have entered non-natal tributaries emigrate back into the main-stem Klamath River. We defined the winter emigration period as November 1–January 31 and relied on estimates of non-natal tributary survival from Manhard and others (2018a). To estimate the abundance of juvenile coho salmon surviving the winter period, we applied a mean annual survival rate (see table 22 in Manhard and others [2018a] for parameter estimates) corrected for the length of time spent in non-natal tributaries calculated as the median non-natal tributary entry date to the end of the winter emigration period. Next, we used the mean proportion of fish that emigrate over the winter period (see table 22 in Manhard and others [2018a] for parameter estimates, winter emigrant proportion) to simulate the number of fish leaving tributaries during the winter versus staying until the spring. Manhard and others (2018a) estimated the timing of fish emigrating from tributaries in relation to environmental covariates. To generalize the emigration timing for all non-natal tributaries in S3, we simplified the approach, relying on a subset of covariates from the lowest AIC model in Manhard and others (2018a). We used the intercept, representing the baseline emigration rate, and the slope term for weekly interval from Manhard and others (2018a) to estimate weekly conditional binomial probabilities. Because the S3 model runs on a daily time-step, we converted the weekly binomial probabilities used in Manhard and others (2018a) to daily, multinomial probabilities across the winter emigration period. As an example of converting from the conditional binomial probabilities to multinomial, given inputs to the binomial such as weekly mean day length representing photoperiod, we plotted an example along with the emigration probabilities for winter and spring periods (fig. 2). The emigration probabilities for winter peak in late December and early January.

Next, we modeled the spring emigration period, defined as February 12–June 30. First, the remaining fish in each non-natal tributary survive (those that have not emigrated during the winter period). These fish are used to predict the abundance of fish available to emigrate during the spring period. We used the mean survival in non-natal tributaries (see table 22 in Manhard and others [2018a], for parameter estimates), corrected for the length of time from the end of the winter period to the end of the spring period. Next, the emigration timing was modeled using a subset of factors identified in the lowest AIC model from Manhard and others (2018a), including an intercept and a slope representing the effect of photoperiod. By using the intercept and photoperiod effect, this relationship can be easily generalized across all non-natal tributaries, allowing for expanding the number of non-natal tributaries in future applications of the model. We converted the predictions from weekly to daily multinomial probabilities across the spring emigration period. For the spring period, figure 2 shows an example of converting from the conditional binomial probabilities, based on inputs (hours of light, fig. 2A, for spring) to the multinomial probabilities. The spring emigration probability peaks in mid-May. Thus, the daily number of fish emigrating to the main-stem Klamath River was simply calculated as the total abundance of fish (calculated by applying the survival) multiplied by the daily emigration probability.

Because growth within non-natal tributaries was not simulated in S3, we estimated the monthly mean length of fish emigrating from tributaries to the main-stem Klamath using a large dataset of coho salmon size-at-date data. These measurements span a range of habitats and are similar to estimated sizes of emigrants from natal tributaries during overlapping periods. These mean monthly sizes of fish are applied to any fish emigrating from a tributary to the main-stem Klamath. Fish enter the Klamath River at their respective tributary mouths and the full set of S3 daily dynamics is then applied during the remainder of a simulation.

Disease Model

Model selection proceeded in three steps for the mixture cure model (see “Disease Modelling” section). First, the generalized F distribution was the most highly supported distribution, based on AIC, for the time until death component of the mixture cure model (table 4). The log-logistic distribution was the next most highly supported model, but was greater than 2 ΔAIC from the generalized F. Second, in the survival timing portion of the model, the logarithm of spores was the most highly supported, by greater than 9 and 14 ΔAIC, from the full and main covariate model, respectively, compared to using just the spore concentration. We used the logarithm of spores in the survival timing portion of the model for further model fitting and selection. Third, we evaluated interactions by removing each individual interaction and only keeping interaction terms that reduced AIC by more than 10 units compared to the full model. This strategy resulted in only one interaction being kept, the exposure temperature and holding temperature interaction. Additionally, we kept the other main effects that included log spores, exposure temperature, holding temperature, and exposure duration.

The most highly supported mixture cure model generally was able to capture important elements of disease mortality dynamics of juvenile coho salmon infected with C. shasta (fig. 3). Comparing the Kaplan-Meier survival curves to the model predictions shows that the most highly supported model captured the delayed onset of mortality, the period of high mortality, and the plateau where mortality rate levels off. The model predicted higher mortality than was observed for several trials (numbers 7, 17, 24) and slightly lower mortality in several more (trial numbers 6, 10, 12). However, in general, the model captured patterns in mortality associated with fish infected with C. shasta. For the cure portion of the model that estimates the probability of infection and eventual death, the coefficients for C. Shasta spore concentration, exposure temperature, and exposure duration were positive (table 5). The coefficients for holding temperature and an interaction between exposure temperature and holding temperature were negative. In the time-to-death portion of the model, all coefficients representing covariate effects were negative, except an interaction term between exposure temperature and holding temperature, which was positive.

We made predictions of survival probability across a range of covariates to illustrate the simultaneous effects of both portions of the mixture cure model on disease progression (fig. 4). Coho salmon are sensitive to even low spore concentrations, and across a range of other covariates, as the concentration of spores increased, the rate and overall mortality increased. The effects of elevated temperature on C. shasta mortality are clear—as temperature increases, mortality increases (moving down rows in fig. 4). Additionally, as exposure time increases, mortality increases across a range of spore concentrations.

Model Inputs

Based on the models of Manhard and others (2018a), the peak of adult migration timing in the Scott River showed considerable variation between years, ranging from mid-October to late December (fig. 5A). Adult coho salmon migrating into the Shasta River showed less variation in timing between years, with the peak of migration typically occurring in mid-November to early December (fig. 5B).

For 7 brood years (2007–13), we calculated spring age-0, fall age-0, and spring age-1 fish entering the Klamath River from the Scott River (fig. 6). The peak for spring age-0 fish generally occurred in May and June, although the peak for spring age-1 fish was more variable, with peak abundance occurring from March through June in some years. Total abundance ranged from 434 in 2009 to 54,403 in 2007 (table 6).

Similarly, we calculated juvenile fish abundance and emigration timing from the Shasta River for 9 brood years from 2005 to 2013 (fig. 7). The peak for spring age-0 fish generally occurred in May and June; however, the peak was later in July during 2009 and 2010. Most spring age-1 fish generally migrated into the Klamath River in April and May. Total abundance of inputs from the Shasta River ranged from 54 in 2009 to 4,670 in 2007 (table 7).

The mean total length across all brood years of fish entering the Klamath River from the Scott River was 56, and 109 mm for spring age-0, and spring age-1, respectively (fig. 8A). For the Shasta River, mean total length of migrating fish was 65, and 134 mm for spring age-0, and spring age-1, respectively (fig. 8B). Variation between years was caused by annual differences in temperature and ration levels that were used in the models to generate inputs for S3. Note the total length of fall age-0 fish was set to 90 mm.

Genotype II spore concentrations were variable among years, ranging from near zero to 50 spores per liter in 2015 (fig. 9). A noticeable peak in spore concentrations occurred in 2007, 2008, 2009, 2014, and 2015 with elevated concentrations typically during April, May, and June. In other years, spore concentrations were lower with the absence of any large peak of elevated concentrations.

Model Output

Tributary use by spring age-0 fish was variable, often depending on brood year and the location of non-natal tributaries (fig. 10A–10B). In general, non-natal tributaries farther upstream received the most use, and the number of fish entering tributaries decreased moving downstream. For Scott River fish, Tom Martin Creek had the highest use and for Shasta River fish, Horse Creek received the most fish. For a given non-natal tributary, the number of fish entering each year showed considerable variation. For instance, Scott River spring age-0 fish use of Tom Martin Creek varied by several thousand fish across brood years. Brood year 2010 had the lowest non-natal tributary use for Scott River fish with only 35 individuals simulated to enter. For Shasta River fish, brood year 2009 had the lowest non-natal tributary use.

Spring age-0 fish generally started entering non-natal tributaries in mid-May, with peak entry generally occurring in late June or early July (fig. 11; appendix 1). For some brood years, the start of entry was much later (mid-June or July). Entry into non-natal tributaries generally decreased by August. The entry timing was often unimodal; however, in several years the distributions of entry timing had multiple peaks.

We present the population level impacts of C. shasta on coho salmon in several ways. First, using the fitted cure model and the S3 disease submodel, we track the daily percentage of infected fish passing Seiad Creek, typically considered the downstream boundary of the infectious zone (fig. 12). The percentage of infected fish showed considerable variation among years, with some years having high levels of infection (2007) and other years having low levels (2009; near 0 percent years omitted from figure). Infected spring age-0 fish started passing Seiad Creek in April and sustained high levels of infection during the summer months with infection rates generally decreasing by September. There were virtually no fall age-0 fish infected. Infected age-1 spring fish were primarily observed from the Shasta River. These fish started to pass Seiad Creek in April, continuing until September. Recall that our use of the term “infected” here refers to fish that become infected and eventually die at some point in time after infection, which is determined by the fitted cure model.

Second, we track the daily percentage of infected fish entering the ocean (fig. 13). For some brood years and life histories, the percentage of infected fish entering the ocean was up to 40 percent (2005, 2007), while other brood years had a much lower percent entering the ocean (2009, years omitted from figure). Infected fish entering the ocean primarily consist of spring age-0 and age-1 life-histories, with virtually zero fall age-0 fish (table 8). Out of the spring age-1 life-history, fish from the Shasta River make up the dominant portion of infected fish entering the ocean.

Third, to examine the overlap in timing between fish leaving the infectious zone and periods of peak spore concentrations, we present a series of paired plots showing the passage at Seiad Creek with the corresponding spore concentrations (fig. 14). In general, the passage timing of spring age-0 and age-1 fish overlapped with periods of elevated spore concentrations. For spring age-0 fish passing Seiad, this typically started in April or May and continued through the first 2 weeks of July. Age-1 fish from a given brood year, generally passed Seiad Creek during a more protracted period, the following calendar year, with a similar peak passage time as age-0 fish. Fall age-0 fish passed Seiad Creek starting in late fall until early spring, a period of low spore concentrations in the infectious zone.

Fourth, we plot the number of in-river mortalities resulting from C. shasta as a function of river kilometer for each of the brood years considered (fig. 15). Mortality occurred across a wide range of river kilometers, however, most mortality occurred between river kilometers 150 and 250. Total number of in-river mortalities ranged from 0 in some years to 122 individuals from brood year 2007.

Lastly, we summarize percentage of mortality from C. shasta for each life-history strategy, source, and brood year (fig. 16). We stratify this percentage of mortality based on the simulated “fate” of infected individuals, including Klamath River, tributary (non-natal) and ocean. Most C. shasta-related mortality occurred in non-natal tributaries, where infected fish entered tributaries and were predicted to die. Infected fish that entered the ocean were the next largest source of C. shasta related mortality, particularly for spring age-1 fish from the Shasta River. These fish are smolts, actively migrating towards the ocean, and once infected in the infectious zone, have short travel times to the ocean. In-river mortality made up the smallest component, mostly affecting spring age-0 fish from the Shasta River. There were large differences in the percentage of mortality between natal tributaries and life histories. Overall, across all life histories, C. shasta mortality was lower in the Scott River than in the Shasta River, especially for spring age-1 fish. In both natal tributaries, fall age-0 fish had near zero mortality compared with spring age-0 or spring age-1 fish.

We simulated the daily number of coho salmon entering the ocean for each brood year and life history (fig. 17). The timing of ocean entry is similar across years for each life-history strategy. Age-0 spring fish generally showed three distinct peaks of ocean entry depending on the migratory pathway taken: (1) fish that remain in the main stem enter the ocean early during their first summer, (2) non-natal tributary use and winter emigration results in entry during winter as age-1 fish, and (3) spring emigration results in typical spring or summer ocean entry as age-1 smolts. Fall age-0 fish typically entered the ocean after a short residence in the main-stem Klamath River during winter from November through February. Spring age-1 fish entered over a longer time from March through June with most ocean entry finished by July.

We summarize these daily ocean entries as totals for each life history from the Scott and Shasta Rivers (fig. 18A–18B). The Scott River had higher abundance of juvenile coho salmon entering the ocean with abundance in the highest individual brood year (2007) topping 45,000 fish (combined across life histories). However, brood year 2007 stands out as having high production of fish, and in most years, abundance entering the ocean was considerably lower. Several hundred to several thousand fish produced from the Scott River was more typical. Spring age-1 fish generally accounted for most of the fish produced from the Scott River with lower numbers for both spring and fall age-0 fish.

The Shasta River produced lower numbers of fish reaching the ocean compared to the Scott River. The highest year for the Shasta River was brood year 2007 with more than 2,400 simulated to enter the ocean. Production of fish from brood year 2009 was particularly low, with only 133 fish simulated to reach the ocean. The number of fish representing each life-history strategy showed less variation in the Shasta River compared to the Scott River.

Given the predictions of total fish entering the Klamath River and the abundance of fish entering the ocean, we calculated the percentage of fish surviving to ocean entry (fig. 19A–19B). These percentages generally reflect differences in life history, given various physical and biological inputs, and the time spent in the main-stem Klamath River and non-natal tributaries. For the Scott River, spring age-0 fish showed the most variation in survival from 13.8 to 50.1 percent, while survival for the other life histories was relatively constant. This pattern also was seen in Shasta River fish where most of the variation in survival was demonstrated by spring age-0 fish. Survival was lowest for spring age-0 fish from the Shasta River of any group considered, with survival to ocean entry ranging from 2.1 to 13.0 percent.

Mean size at ocean entry was tracked in simulations, and boxplots of fork length of coho salmon smolts at ocean entry are shown in figure 20. The largest smolts at ocean entry were spring age-1 fish from the Shasta River. Spring age-1 fish from the Scott River were smaller and the mean size overlapped with the other life-history strategies in the Scott River. Spring age-0 fish from both rivers had considerable variation in the mean size at ocean entry with some fish entering around 90 mm fork length, whereas others entered close to 135 mm. Fall age-0 migrants showed little variation in size at ocean entry.

We explore the effects of density dependent movement resulting from high densities of juvenile Chinook salmon in the main-stem Klamath River. To illustrate the effects, daily number of spring age-0 fish passing Seiad Creek were plotted with and without density dependent movement (Scott River, fig. 21; Shasta River, fig. 22). The effects of Chinook salmon densities generally were minor on the passage time at Seiad Creek.

Total abundance of spring age-0 coho salmon entering the ocean for each brood year were simulated with and without Chinook salmon (fig. 23). The simulation with Chinook salmon resulted in fewer juvenile coho salmon reaching the ocean in all brood years. Generally, the differences in the number of coho salmon reaching the ocean with and without Chinook salmon were small for the Scott River source. There were some brood years (2005, 2011, 2012) where fish from the Shasta River more than doubled when Chinook salmon were removed from the simulations. The percentage of spring age-0 surviving to ocean entry also were simulated with and without Chinook salmon (fig. 24). The percentage of Scott River spring age-0 fish surviving to ocean entry showed similar patterns to the number entering the ocean. Shasta River spring age-0 fish showed the most variation among simulations with and without Chinook salmon. Without Chinook salmon, survival of Shasta River fish was higher, around 12 to almost 20 percent across years, compared with survival of less than 15 percent with Chinook salmon. In comparison, the survival to ocean entry was more similar for spring age-0 fish from the Scott River, although survival without Chinook salmon was higher in all years. The size of coho salmon smolts from both natal sources entering the ocean were similar with and without Chinook salmon in the simulations (fig. 25).

Next, we compared the non-natal tributary use with and without Chinook salmon (figs. 26A–26B). For coho salmon from the Scott River, the non-natal tributary use without Chinook salmon was shifted downstream to some extent with more overall non-natal tributaries being used (compare fig. 26A with fig. 10A). Generally, the use of these tributaries by coho salmon increased without Chinook salmon, with higher total abundance predicted to enter non-natal tributaries. For fish from the Shasta River, non-natal tributary use without Chinook salmon shifted downstream slightly and the abundance of fish entering non-natal tributaries generally increased (compare fig. 26B with fig. 10B). The increased use of non-natal tributaries without Chinook salmon was most pronounced at upstream locations, particularly Horse and Tom Martin Creeks.