Population Trends for Chinook and Summer Chum Salmon in Two Yukon River Tributaries in Alaska

All five species of Pacific salmon (Oncorhynchus spp.) common to North America make spawning runs up the Yukon River (Groot and Margolis 1991), the largest drainage in Alaska and the fifth largest in North America (Revenga et al. 1998). Chinook O. tshawytscha and Chum Salmon O. keta are the most widely distributed species in the river and support significant fisheries in the U.S. and Canadian sections of the drainage (Bue et al. 2009; Evenson et al. 2009). Yukon River Chinook Salmon are considered to be ‘stream type' because nearly all of the young remain in freshwater for one full year after emergence before migrating to sea (Healy 1991; Murphy et al. 2017). Age 0 and age 2 freshwater Chinook Salmon are rare and often absent in annual age collections within the drainage (e.g., DuBois and Liller 2010; Horne-Brine et al. 2011). Most smolts arrive in the estuary during June and July (Hillgruber and Zimmerman 2009). Their marine distribution is restricted to the Bering Sea and nearby regions of the North Pacific Ocean just south of the Aleutian Islands (Myers et al. 2009; Larson et al. 2013). Over 99% of Yukon River Chinook Salmon return to spawn at brood year ages 4–7, with ages 5 and 6 being the dominant age classes most years (Healy 1991; Evenson et al. 2009; Estensen et al. 2012).

Chum Salmon return to the Yukon River as two distinct groups or races referred to as summer and fall Chum Salmon (Buklis and Barton 1984; Salo 1991; Flannery et al. 2007). Summer Chum Salmon return in early to midsummer and spawn in main-stem reaches of runoff streams in the lower Yukon River. Few summer Chum Salmon ascend beyond 1,600 km upstream in the Yukon River or its tributaries. By contrast, fall Chum Salmon return in late summer and spawn in late fall or early winter in upper Yukon River tributaries or main-stem regions with perennial springs or upwelling areas (Bue et al. 2009; Fleischman and Borba 2009; Wirth et al. 2012). Juvenile Chum Salmon migrate downstream to the sea the summer following emergence and most arrive in the estuary between mid-June and mid-July (Hillgruber and Zimmerman 2009; Vega et al. 2017). Yukon River Chum Salmon migrate through the Bering Sea and into the North Pacific Ocean by their first winter at sea (Salo 1991; Myers et al. 2009; Urawa et al. 2009). They feed primarily in the North Pacific Ocean until the spring in which they mature and migrate back to the Yukon River. Long-term aging data from commercial and subsistence harvests suggest that summer Chum Salmon returns in the Yukon River are dominated by brood year ages 4 and 5 in nearly equal measures, with very few fish returning at age 3 or 6 (Estensen et al. 2012). Fall Chum Salmon returns are strongly dominated by age 4 fish, almost always >50%, with progressively smaller components of age 5, age 3, and age 6 fish (Bue et al. 2009; Fleischman and Borba 2009; Estensen et al. 2012).

Yukon River Chinook Salmon and both races of Chum Salmon are subject to substantial commercial, subsistence, and other fisheries as they migrate up the river to their natal spawning streams each year (Estensen et al. 2012). These fisheries are managed to achieve various escapement or fish passage objectives, provide opportunity for subsistence and other personal use fisheries, and allow commercial fishery harvests, in that priority order (Bue et al. 2009; Evenson et al. 2009; Estensen et al. 2018). The decision to open or close fisheries during the fishing season is based primarily on run strength indicators in the lower reaches of the main-stem Yukon River, where fish from all upstream spawning stocks migrate together. These indicators include test fisheries, which are set or drift gillnets operated systematically to provide daily measures of relative abundance for the multiple fish species in the river, the Pilot Station sonar project, which combines a test fishery with a sonar counting project to provide quantitative estimates of abundance (Lozori and McIntosh 2014; Pfisterer et al. 2017), and genetics projects that apportion fish passage among upstream spawning stocks (e.g., Smith et al. 2005; Flannery et al. 2007; Beacham et al. 2008). In addition to these lower main-stem abundance indicators, there are several projects in tributary streams that collect annual escapement, run timing, and demographic data from specific spawning populations (e.g., Savereide and Huang 2014; Mears and Morella 2017; Wilson 2017). These tributary projects occasionally are used by fishery managers during the season to corroborate run strength qualities observed in the main stem, assuming general trends would be similar, but they are primarily used as postseason assessments of management actions.

In addition to major fisheries, Chinook and summer Chum Salmon populations have experienced significant environmental changes in both marine and freshwater habitats in recent years. There is substantial annual variability, but marine habitats have been trending warmer in recent years, sea ice cover and duration have been declining, and fish and invertebrate communities that prefer warmer conditions have been expanding north displacing those that prefer colder conditions (Mueter and Litzow 2008; Royer and Grosch 2009; Stabeno et al. 2017). The warming marine conditions in the Bering Sea have been shown to increase growth rates of rearing Chinook Salmon, resulting in earlier maturation and a younger overall spawning population (Siegel et al. 2017, 2018). In the freshwater environment, rivers and lakes are now experiencing measurably earlier breakup dates and later freeze-up dates than occurred 100–150 y ago (Magnuson et al. 2000). Breakup in the Yukon River, for example, takes place approximately a week earlier now than it did in 1900 (Brabets and Walvoord 2009). Mundy and Evenson (2011) developed an empirical model describing average spawning migration timing of adult Chinook Salmon into the Yukon River from 1961 to 2009 as a factor of sea surface temperature, air temperature, and sea ice cover in the eastern Bering Sea. Their model predicts that the spawning migration should take place earlier as the environment warms. It is likely that these types of natural environmental changes are influencing other aspects of salmon biology and life history as well.

There is a growing interest among researchers and stakeholders in the Yukon River in evaluating Chinook Salmon runs for measures of quality rather than simply numbers of fish (Schindler et al. 2013). It is generally understood that the quality of a Chinook Salmon run is enhanced when it includes larger rather than smaller fish, older rather than younger fish, and a greater rather than lesser proportion of females (Schindler et al. 2013; Hixon et al. 2014). Among salmon species in the Yukon River, Chinook Salmon have the widest range of sizes, from approximately 450 mm to >1,000 mm mid-eye to tail fork (MEF; e.g., Schuman and DuBois 2011; Larson 2019), with most small and young fish being male and most large and old fish being female (Healey 1991). These qualities makes the species vulnerable to having one component of the population more heavily exploited than another because of the size-selective nature of gillnet fisheries (Bromaghin 2005). Summer Chum Salmon are less vulnerable to size-selective fisheries because the species has a much narrower length distribution, from approximately 450 mm to 650 mm MEF, with nearly equal fractions of males and females in both of the dominant age classes, and mean lengths of males and females within age classes differing by <30 mm in most collections (e.g., Schuman and DuBois 2011; Larson 2019). The Chinook Salmon run in the Yukon River coincides with the much more numerous summer Chum Salmon run (Eiler et al. 2014; Larson et al. 2017), so fisheries directed for Chinook Salmon have most often used large mesh gillnets that select for large Chinook Salmon and allow summer Chum Salmon and small Chinook Salmon to pass through the nets without being captured (Howard et al. 2009; Howard and Evenson 2010). The immediate consequence of this type of fishery is that the largest individuals in the Chinook Salmon run are more heavily exploited, presumably leaving escapements with smaller and younger fish of both sexes and fewer females. Some recent literature indicates declining length at age over the past several decades for Chinook Salmon from several North Pacific river systems, including the Yukon River (Bigler et al. 1996; Lewis et al. 2015; Ohlberger et al. 2018). Size and age at maturity for Chinook Salmon have been shown to be heritable qualities (Hankin et al. 1993; Heath et al. 1994; Berejikian et al. 2011); therefore, the long-term consequences of this type of selective fishery are thought to include an evolutionary decline in the proportion of large and old females returning to spawn, a reduction of the average size and age of returning fish, and a reduction in the productivity of the population (Hard et al. 2009; Bromaghin et al. 2011b).

In this manuscript we examine trends over time in escapement (returning fish that successfully reach their spawning destinations), run timing, demographic composition, and production of Chinook and summer Chum Salmon (hereafter referred to as Chum Salmon) populations migrating past weirs located in the East Fork Andreafsky (EF Andreafsky) and Gisasa rivers. Weirs in both rivers have been operated by the U.S. Fish and Wildlife Service (USFWS) since 1994 (Carlson 2017; Mears and Morella 2017) and trends observed here provide insight into population-level response during a period of significant changes in drainage-wide production, environmental conditions, and harvest practices. In addition to describing various population characteristics of the two Chinook and Chum Salmon populations, we specifically address four key questions: 1) are annual indicators of run strength from the main-stem sonar project at Pilot Station, as revised by Pfisterer et al. (2017), positively correlated with weir passage; 2) is run timing trending earlier over the observed time frame as predicted for a warming environment based on modeling work by Mundy and Evenson (2011); 3) have there been changes over time in age composition, sex ratio, and length-at-age consistent with findings in other North American Pacific salmon populations assessed at broader spatial scales (Bigler et al. 1996; Lewis et al. 2015; Ohlberger et al. 2018); and 4) have there been changes in productivity that we might expect if smaller and younger fish become more common in the spawning streams (Schindler et al. 2013; Hixon et al. 2014)?

The Yukon River basin is the largest in Alaska and the fifth largest by drainage area in North America behind the Mississippi, Mackenzie, Nelson, and Saint Lawrence rivers (Revenga et al. 1998). It drains an area of approximately 850,000 km², approximately 500,000 km² of which is in Alaska (Brabets et al. 2000). The Yukon River flows >3,000 km from its headwaters in northern British Columbia, Canada, to its mouth at the Bering Sea. Average annual flow near the Yukon River mouth is approximately 6,400 m³/s, and peak flow in early summer averages approximately 20,000 m³/s.

In this manuscript we focus on two lower Yukon River tributaries (Figure 1) that have had fish weirs in place since 1994: the EF Andreafsky River in the Yukon Delta National Wildlife Refuge (NWR; Mears and Morella 2017), and the Gisasa River in the Koyukuk NWR (Carlson 2017). The EF Andreafsky River is a sixth-order stream (Strahler 1957) that flows in a south-southwest direction from its headwaters in the southern Nulato Hills to its confluence with the main-stem Andreafsky River, approximately 8 km from the mouth of the Andreafsky River and 175 km from the south mouth of the Yukon River, which is the largest distributary in the Yukon River delta (McDowell et al. 1987). The EF Andreafsky River drains an area of approximately 1,970 km² and the valley is approximately 170 km long. The Gisasa River is a fourth-order stream that flows northeast from its headwaters in the northern Nulato Hills to its confluence with the Koyukuk River, approximately 90 km from the mouth of the Koyukuk River and 908 km from the mouth of the Yukon River. The Gisasa River drains an area of approximately 1,292 km² and the valley is approximately 115 km long. Both drainages are in the interface between the continental climate common to interior Alaska and the maritime climate common to the Bering Sea coast in western Alaska (Shulski and Wendler 2007). Air temperatures average approximately 17°C to 20°C in midsummer, with extreme high temperatures reaching 30°C or warmer, and −20°C to −25°C in midwinter, with extreme low temperatures of −50°C or colder. Annual precipitation averages from 30 cm to 38 cm. Both rivers generally start the freeze-up process in October and break-up in late April or early May.

The USFWS has deployed and operated resistance board weirs, as described by Tobin (1994), in the EF Andreafsky and Gisasa rivers every summer from 1994 to 2016 (Carlson 2017; Mears and Morella 2017). The EF Andreafsky River weir was originally installed at latitude 62.12996° and longitude −162.79793° (WGS84 datum; Tobin and Harper 1995), approximately 40 km upstream from the confluence with the main stem. It was moved in 1995 to its current location, approximately 2 km downstream, at latitude 62.11673° and longitude −162.80761°, which had a wider and shallower channel profile thought to be better able to handle high flow periods (Tobin and Harper 1996; Mears and Morella 2017). The weir spans a distance of 105 m and has had two fish passage chutes since the 1996 season (Tobin and Harper 1997). Count start dates ranged from 15 June to 2 July, depending on flow levels, with the exception of 2001 when high water prevented counting until 15 July (Zabkar and Harper 2003). Count stop dates ranged from 27 July to 3 August, based on diminishing counts of Chinook and Chum Salmon, during the years 1994 and 2006 to 2016. During the years 1995 to 2005, USFWS researchers made an effort to count later arriving Coho Salmon O. kisutch and the weir remained operational until a fall stop date that ranged from 11 September to 23 September.

The Gisasa River weir has been installed in the same location every year, approximately 4.5 km upstream from the mouth, at latitude 65.25254° and longitude −157.71561°. The weir spans 45 m and has one fish passage chute (Carlson 2017). Count start dates ranged from 20 June to 29 June with the exception of 2001 when high flow prevented counting until 7 July (VanHatten 2002). The USFWS determined count stop dates following 3 consecutive days in which fish counts for both Chinook and Chum Salmon were <1% of the seasonal passage at that time and ranged from 27 July to 8 August (Carlson 2017).

An observer counted all fish migrating through the fish passage chutes. Prior to 2014, the observer sat on the chute counting fish whenever the doors on the chute were open. Beginning in 2014, motion-activated, high-resolution video systems were developed and since that time fish have been counted through both weirs by observing video images of migrating fish (Carlson 2015; Mears 2015). Observers specifically identified most species as they swam through the chutes, but they tallied coregonid species simply as members of the subfamily Coregoninae.

The USFWS sampled Chinook and Chum Salmon for age, sex, and length data during each week of weir operations with a few weather-related exceptions. In the early years of the EF Andreafsky (Tobin and Harper 1996) and Gisasa (Wiswar 2000) River weirs, researchers made an effort to achieve weekly sampling goals during a 1–4-d period at the start of each weekly stratum to maximize demographic contrasts among strata, a strategy discussed by Geiger et al. (1990). Later, both weir projects adopted a daily sampling plan with effort distributed across different hourly time periods each day (Carlson 2017; Mears and Morella 2017). Sampling consisted of closing the upstream door in the trap and waiting until a desired number of salmon of the target species entered the trap and then closing the downstream door. Researchers then sampled all individuals of the target species and released them upstream from the weir. Neither sampled fish nor those simply being counted migrating through the weir were able to migrate back downstream. Thus, individual fish were available for counting and sampling only once.

Researchers placed fish in a cradle to restrict their movements in preparation for sampling. Researchers measured each fish mid-eye to tail fork (MEF) to the nearest 5 mm and classified it as either male or female based on external morphology. Researchers collected scales for aging from the left side of sampled fish, two rows above the lateral line on a diagonal from the posterior insertion of the dorsal fin to the anterior insertion of the anal fin (Koo 1962; Devries and Frie 1996). Researchers collected three scales from Chinook Salmon and one scale from Chum Salmon. Researchers mounted these scales on labeled gum cards and submitted them to the Alaska Department of Fish and Game aging lab for analysis. These demographic sampling data are available in Table S1 (Supplemental Material).

Age estimates are generally reported in European notation, in which the winters in freshwater after hatching and the winters in saltwater are identified and separated by a period (Groot and Margolis 1991). Salmon hatch and emerge in late winter or spring following the spawning event, so the brood year age for a salmon becomes the sum of these two numbers plus one. For example, an age 1.3 Chinook Salmon is brood year age 5, and an age 0.4 Chum Salmon is also brood year age 5. With very few exceptions, Chinook Salmon in the Yukon River are freshwater age 1 (Healy 1991; DuBois and Liller 2010; Horne-Brine et al. 2011) and Chum Salmon are freshwater age 0, so we simplify analysis and reporting here by using brood year ages.

Ideally, all fish passing the weirs are observed and counted; therefore, the sum of daily counts during a season should be a census of spawning salmon (Parsons and Skalski 2010). However, there were often small numbers of fish present the first day a weir was operational, indicating that a small fraction of the run migrated upstream early. Similarly, the extended seasons of the EF Andreafsky River weir during the years 1995 through 2005 revealed that a small fraction of the Chinook and Chum Salmon runs continue trickling in for days or weeks after the main body of the run passes. Additionally, during some years, high flow events and associated turbidity following large rains have interfered with counting for a few hours to a few days during the runs. Over the course of these weir projects, the early and late gaps in counts have generally been considered insignificant and were not reconstructed. However, the midseason gaps were estimated in a variety of ways, including proportionally expanding partial day counts (Carlson 2015), linear interpolation of passage between the day before to the day after a missed segment of one or more days (Melegari 2011), and using the history of run proportions from specific date ranges in other years to reconstruct gaps in the current year's count record (Zabkar and Harper 2003).

We investigated trends in abundance for populations within each tributary and correlations between populations. We used Negative binomial regression models to test for trends in abundance for each population. We also used Spearman's rank correlation to test each species for the strength of relationships among tributary escapement estimates and the passage estimates from the Pilot Station sonar program, which is the most quantitative of the run-strength indicators in the lower main-stem Yukon River (Lozori and McIntosh 2014; Pfisterer et al. 2017). Where relationships were suspected, we fitted a Negative binomial regression model to the data to characterize this relationship. We explored the sensitivity of the estimated slope to error in the independent variable by refitting the regression with simulated data sampled from the estimated distributions of the independent data, assuming normality.

All models throughout this paper were fit using the Program R statistical package (R Core Team 2018). Unless otherwise noted, we performed model selection using Akaike's Information Criteria corrected for small sample sizes (AIC_c; Burnham and Anderson 2002) and reported parameter estimates with 95% confidence intervals or standard errors. We checked final models for assumptions of normality, constant variance, and independence with appropriate residual plots. Where necessary, we added autoregressive terms to models to address nonindependence. We assessed model fit by comparing simulated data (or residuals) generated from the final model with observed data (or residuals), in an approach described by Zuur et al. (2009).

Although our time series on the EF Andreafsky and Gisasa River weirs is only 23 y, we tested the null hypothesis that run timing of the Chinook and Chum Salmon populations was not trending earlier over this time period. Recognizing that the days in which the first and last fish of the annual salmon runs arrived in the EF Andreafsky and Gisasa rivers were not empirically determined for all years, we described annual timing of Chinook and Chum Salmon runs by identifying the non-leap-year Julian days in which the first quartile, midpoint, and third quartile of each annual spawning run was observed. We fitted a linear regression model to each quantile as a function of time to determine whether an association between this phenology trait and time existed. We also included tributary passage estimates, tributary identity, and the interaction between time and tributary in the full models to determine whether run timing was associated with escapement level or whether differences in run timing between EF Andreafsky and Gisasa rivers were consistent over time. We investigated the sensitivity of the results to the addition of interpolated counts by refitting the model to the raw data.

Examining age structure of the escapement can be confounded by the variable brood-year cohort strengths making up that population, and for that reason brood-year age structure may be preferable if trends in age structure over time is the focus of the analysis (Lewis et al. 2015). However, estimating brood-year age structure requires estimation of the harvest component of the run and its age structure, both requiring assumptions as described in the spawner recruit analysis below. Therefore, we chose to focus the analysis of age structure on the return year because those data were directly observable at the weirs.

Using formulae outlined by Thompson (1992; see Text S1, Supplemental Material) we estimated the proportion of each age class in annual escapements with associated variances by poststratifying the sampled fish into three strata of similar magnitude weighted by the escapement proportion within each strata. We first considered trends in the mean annual age of the escapements, calculated as a weighted average of the four dominant age classes represented in each population using formulae presented by Taylor (1997; see Text S1). We used a least-squares linear regression, weighting observations by the inverse of their variances, to test the null hypothesis that the slope parameter of mean annual age of the escapement by year was equivalent to zero for each of the four populations.

Chinook Salmon are more profoundly affected by size-selective fisheries than are Chum Salmon, as discussed earlier; therefore, we considered the alternative hypothesis that mean annual age of Chinook Salmon escapements could increase significantly during our time series because of major changes in the fishery. Harvest levels in the lower Yukon River, for example, were much greater during the first few years of our time series when the commercial fishery was operating fully, and declined dramatically during the last few years (Evenson et al. 2009; Estensen et al. 2018). In addition, large mesh gillnets, mostly 21.6-cm stretch mesh, were used to harvest Chinook Salmon during the early and middle periods of our time series until the commercial fishery was eventually closed in 2008, a regulation was established in 2010 that specified a maximum allowable stretch mesh size of 19 cm in the drainage, and subsistence harvest declined to negligible levels between 2013 and 2016. We reasoned that if mean age was more profoundly influenced by the selective fishery than by large-scale environmental effects (Siegel et al. 2017, 2018), then the mean age of Chinook Salmon escapements should increase over time.

During the 2007 season, we made a procedural decision to classify all age 4 Chinook Salmon from weir samples as male (Maschmann 2008; Melegari 2008). We made this decision because the female proportions of age 4 fish reported at the weirs, where sex was classified based on external morphology, were anomalously high when compared with similar data from sampling projects where sex was classified based on internal examination of the gonads or projects in which fish were sampled farther upstream in the drainage where sex-specific morphology was more developed (Bales 2007; Karpovich and DuBois 2007). Karpovich and DuBois (2007) specifically pointed out this anomaly in age 4 data from the EF Andreafsky River weir. We reexamined this 2007 decision, considering large samples of Chinook Salmon from main-stem collection locations in which sex was classified based on internal examination of gonads, and confirmed the basis for this decision (additional details in Text S2, Supplemental Material). We subsequently classified all age 4 Chinook Salmon as male for the entire data set. We fit age structure and length-at-age models with and without sex parameters to ensure that our inferences about trends over time were not sensitive to potentially incorrect sex classifications.

We fit logistic-binomial regressions using the logit link function to the count data for four categories of age for Chinook (≤4, 5, 6, and ≥7) and Chum Salmon (3, 4, 5, and ≥6) to describe changes in the proportion of each age class over time. Tributary identity and its interaction with year were included in the full model to test for differing trends in EF Andreafsky and Gisasa River populations. For age 5 and 6 Chinook Salmon, where large fractions of both male and female fish were observed, we also included an additive effect of sex in the full model. In all cases counts were overdispersed. The quasibinomial, an approximation of a beta-binomial distribution (Gelman and Hill 2007), was assumed so that we could obtain appropriate standard errors and confidence intervals for the parameter estimates and we used the quasi-Akaike Information Criteria (Lebreton et al. 1992) to test for the retention of covariate terms.

In a similar manner to age composition estimates, we used formulae outlined by Thompson (1992; see Text S1) to estimate mean length at age in annual escapements with associated variances of male and female Chinook and Chum Salmon. We used the same stratification and weighting procedures used for age structure analysis to analyze length at age. We fit weighted linear regressions to each time series, where weights were the inverse of the estimated variance of the mean length. We restricted regression analyses to ages 4, 5, and 6 because annual sample sizes were often small (n <5) for the other age classes. Full models included year, tributary, a year × tributary interaction, and an additive effect of sex where appropriate.

We conducted comparisons between the two spawning streams of overall mean lengths × age and sex categories using weighted means, as detailed by Taylor (1997; see Text S1). We used 2-sample t-tests of null hypotheses that overall weighted mean MEF lengths were similar among streams for each age and sex category of Chinook and Chum Salmon. We evaluated significance in these tests considering a family-level Bonferroni adjustment to α (Verhoeven et al. 2005) for individual contrasts within species: six contrasts for Chinook Salmon (contrast α = 0.008); eight contrasts for Chum Salmon (contrast α = 0.006).

Data related to population production, which is the number of mature recruits per spawner returning to the natal river, including those that are harvested downstream from the natal river and those that escape through the weir, were compiled from several sources. These included weir escapement data, with error as described above, Pilot Station passage data with error as updated by Pfisterer et al. (2017), and harvest data from commercial, test, and subsistence fisheries as tabulated by Estensen et al. (2012, 2018). We considered Chinook Salmon harvest numbers to be without error because they were derived overwhelmingly from commercial harvests with records of all sales. Chum Salmon harvest numbers were derived from a broader mix of commercial and subsistence harvests; therefore, we assigned a prior error of coefficient of variation (CV) = 0.1 to account for uncertainty in those records. We estimated age class proportions in escapements from weir samples with error, as described above. For Chinook Salmon, we took age structure of the harvests in the area downstream from the mouths of the EF Andreafsky and Gisasa rivers, respectively, from a series of annual reports on the origins of Chinook Salmon harvested in the Alaska portion of the Yukon River drainage (e.g., Lingnau 2000; DuBois and DeCovich 2008; DuBois 2018). In these reports, Chinook Salmon harvests in each region of the river originally were classified as being from lower river, middle river, or upper river populations based on scale pattern analyses (Wilcock and McBride 1983; Bromaghin and Bruden 1999) and later using genetics mixed stock analyses (DuBois et al. 2009). Age composition of the harvests were associated with the three different regions of origin and we used the proportional age structure for harvests with origins in the lower Yukon River region as an approximation for harvests from the EF Andreafsky, H(efa), and Gisasa, H(gis), River populations, in a process similar to that used by Neuswanger et al. (2015) and Siegel et al. (2017). For Chum Salmon, we assumed that the harvest was composed of a similar age composition as estimated each year from samples at the two weirs. These escapement year age structure data are available in Table S3 (Supplemental Material).

These brood-year production data are available in a Supplemental Material file at Table S4.

We illustrated the uncertainty in parameter estimates by creating approximate upper and lower 68% confidence interval (CI) curves in a process discussed by Hilborn and Walters (1992), by making a one standard error (SE) increase in the ln(α) parameter while decreasing the β parameter by one SE for an upper bound curve, and the opposite for a lower bound curve. We examined trends in productivity by regressing model residuals over time. Although coarse, the estimates of the errors produced illustrate the imprecision of the biological attributes calculated from these data.

The average annual Chinook Salmon escapement in the EF Andreafsky and Gisasa rivers was 4,278 (n = 22; range = 1,576–8,046) and 2,260 fish (n = 21, range = 1,128–4,161), respectively. No trend in escapement over time was apparent for the population in the EF Andreafsky River; however, the population in the Gisasa River declined at an annual rate of approximately 4.5% (95% CI = 2.9–6.0%) over the observed time series. There was no evidence of any significant correlation among escapements from the two tributary populations and the passage estimates from Pilot Station sonar (Figure 2; Pilot Station and EF Andreafsky River: ρ = 0.37; Pilot Station and Gisasa River: ρ = 0.37; EF Andreafsky and Gisasa rivers: ρ = 0.16; P > 0.10 in each case).

Table 1 details the average and range of dates for the 25th, 50th, and 75th percentiles of the Chinook Salmon runs in the EF Andreafsky and Gisasa rivers. Patterns over time were similar regardless of which percentile was examined; our analyses here focus on the midpoints of the annual runs. In model checking, there was no evidence of a difference in trend over time between the tributaries, but an autoregressive term was required to account for positively autocorrelated errors (φ = 0.48). The pooled estimated slope was slightly positive, but insignificant, a result inconsistent with the hypothesis of earlier runs over time (Figure 3; β_year = 0.028, 95% CI = −0.249–0.305). This result was not sensitive to the inclusion of passage estimates as a covariate in the model or the use of interpolated data. Over the time period examined, the average midpoint of the run in the Gisasa River was 3.78 d (95% CI = 0.01–7.54) later than the run in the EF Andreafsky River.

The proportion of an age class observed in the return was highly variable from year to year, with coefficients of variation as high as 64%. However, the average age of Chinook Salmon in the escapement declined in both EF Andreafsky and Gisasa rivers (β_year = −0.014, 95% CI = −0.022 to −0.006) over the observed time frame (Figure 4). Specifically, the probability of an age 4 fish increased each year by a factor of 1.03 (95% CI = 0.99–1.07), declined modestly for age 5 and 6 fish, and most strongly in age 7 fish (0.94, 95% CI = 0.84–0.96). There were no differences in these observed trends between tributaries.

Average MEF lengths of age 5 and age 6 female Chinook Salmon were larger than respective males in both tributaries (Figure 5; Table 2). Comparisons within age and sex categories between the two tributary populations revealed significantly different average MEF lengths for three of the six contrasts (Table 2), but the actual differences were small, all <15 mm. Chinook Salmon length-at-age trended slightly negative for all age classes: −0.1 mm/y for age 4 fish, −0.7 mm/y for age 5 fish, and −0.4 mm/y for age 6 fish (Figure 5). Only the trend in age 5 fish was statistically different from zero (95% CI = −1.3 mm to −0.1 mm), equating to a predicted decline of 16.3 mm (95% CI = 2.3–30.6 mm) over the 23-y observation period. There was no evidence of a difference in trend by sex or tributary.

Annual brood-year production estimates for Chinook Salmon populations in the EF Andreafsky and Gisasa rivers were significantly correlated (Pearson's r = 0.698, P = 0.003) with similar trends of low and high production (Figure 6). The 2000 brood year produced the highest recruit per spawner ratios of our time series for both populations, 4.6 in the EF Andreafsky River and 1.9 in the Gisasa River. Estimates of exploitation rates for the EF Andreafsky and Gisasa River populations, respectively, ranged from 0.33 (SE = 0.11) and 0.40 (SE = 0.17) during the first 7 y of our time series, to 0.04 (SE = 0.06) and 0.08 (SE = 0.11) during the last 7 y. Data from the EF Andreafsky River population produced a classic Ricker spawner recruit curve (Figure 7). There was a significant linear relationship between ln(R/S) and the number of spawners (F_1,15 = 18.05; P = 0.001; R² = 55%) with parameter estimates for the y-intercept and slope values estimated as ln(α) = 1.367 (SE = 0.354) and β = 3.28 × 10⁻⁴ (SE = 7.7 × 10⁻⁵), respectively. The α-parameter was 3.92, S_MSY was estimated to be 1,687 spawners (68% CI = 1,074–2,605), and maximum recruitment (1/β) was predicted to occur with approximately 3,051 spawners. The sustainable escapement goal for the Chinook Salmon population in the EF Andreafsky River is currently 2,100–4,900 spawners (Estensen et al. 2018), which intersects the upper portion of our estimated S_MSY range. Paired dashed curves illustrated the approximate 68% CIs resulting from parameter estimate uncertainty (Figure 7). Spawner numbers in our data set ranged from 1,576 to 8,046, a contrast of 5.1, while the number of recruits ranged from 1,187 to 8,667, a contrast of 7.3. There was no significant trend in production anomalies over time (Figure 6; β_year = −0.024, 95% CI = −0.084–0.036).

Data from the Gisasa River population fit with the Ricker spawner recruit model produced an arc that never exceeded the replacement line and failed to reach spawner numbers in which expected density-dependent processes would bring recruitment into decline (Ricker 1975). There was no indication of a linear relationship between ln(R/S) and the number of spawners (F_1,16 = 0.02; P = 0.879; R² < 0.17%). Parameter estimates for the y-intercept and slope values were ln(α) = −0.030 (SE = 0.349) and β = 2.1 × 10⁻⁵ (SE = 1.36 × 10⁻⁴). The α-parameter was 0.970 and we did not estimate S_MSY or other population parameters. Spawner numbers in our data set ranged from 1,532 to 4,161, a contrast of 2.7, while the number of recruits ranged from 749 to 4,576, a contrast of 6.1.

Chum Salmon escapements were highly variable but comparable between the EF Andreafsky (n = 22; mean = 68,356; range = 8,886–211,408) and the Gisasa (n = 21; mean = 70,353; range = 10,230–264,690) rivers. No trends in escapement over the observed time frame were detected. Unlike Chinook salmon runs in these rivers, however, annual Chum Salmon escapements were significantly correlated with each other and passage estimates at the Pilot Station sonar (Figure 8; Pilot Station and EF Andreafsky River: ρ = 0.49; Pilot Station and Gisasa River: ρ = 0.93; EF Andreafsky and Gisasa rivers: ρ = 0.61; P < 0.05 in each case). Regression models indicate that an estimated increase of 100,000 fish past the Pilot Station sonar is associated with a 3.5% (95% CI = 1.2–6.0%) increase in the escapement to the EF Andreafsky River and an 8.6% (95% CI = 6.7–10.6%) increase in the escapement to the Gisasa River. By simulating 5,000 data sets using the uncertainties associated with the population estimates and refitting the regression, we found that point estimates for the slope of the regression were insensitive to these uncertainties, all falling within the confidence interval estimated with the observed data.

Table 3 details the average and range of dates for the 25th, 50th, and 75th percentiles of the Chum Salmon runs in the EF Andreafsky and Gisasa rivers. Consistent with the results for Chinook Salmon, there was no support for a linear relationship between year and the midpoint of the run in either tributary system (Figure 9; β_year = 0.013, 95% CI = −0.132–0.159). This result was not sensitive to the inclusion of passage estimates as a covariate in the model or the use of interpolated data. Over the time period examined, the average midpoint of the Chum Salmon run in the Gisasa River was 5.34 d (95% CI = 3.46–7.23) later than the run in the EF Andreafsky River.

Our data indicated that Chum Salmon escapements in the EF Andreafsky and Gisasa rivers are dominated by age 4 and age 5 fish, which together comprised an estimated 95.3% (SE = 1.2%) of the escapement each year. These two age classes often alternated strongly on an annual or longer time period, with proportional contributions sometimes exceeding 0.75 followed by declines to ≤0.25. With highly variable age composition annually, driven primarily by alternating age 4 and age 5 components of the runs, no significant trends over time were apparent in the average annual age of either population (Figure 10; β_year = −0.009, 95% CI = −0.026–0.007).

Data exploration and model fitting supported the inclusion of a quadratic effect of year in the linear models for age 4 (⁠ = −0.105, SE = 0.020) and age 5 (⁠ = −0.058, SE = 0.024) length-at-age. At these ages, across sex and tributary, predicted length-at-age peaked in 2002 (Figure 11). There were no significant trends over time observed by sex or tributary for age 6 Chum Salmon. However, across all ages, for both males and females, fish in the Gisasa River were significantly larger than those in the EF Andreafsky River (Figure 11; Table 4).

Annual brood-year production estimates for Chum Salmon populations in the EF Andreafsky and Gisasa rivers were significantly correlated (Pearson's r = 0.720, P = 0.001) with similar periods of low production, near 1 recruit/spawner or less, and subsequent periods of higher production (Figure 12). The 2001 brood year produced the highest recruit-per-spawner ratios of our time series for both populations, 5.5 in the EF Andreafsky River and 23.7 in the Gisasa River. Estimates of exploitation rates for the EF Andreafsky and Gisasa River populations, respectively, ranged from 0.06 (SE = 0.01) and 0.10 (SE = 0.01) during the first 7 y of our time series, to 0.12 (SE = 0.01) and 0.17 (SE = 0.02) during the last 7 y. The Ricker spawner recruit curves were similar for the two populations when the one extreme outlier from the Gisasa River (2001, recruits = 433,923) was censored (Figure 13). The censored 2001 outlier point was considered to be valid, but it had a substantial influence on the resulting curve. We, therefore, considered the calculations with the outlier censored to be more representative of the underlying relationship between the number of spawners and subsequent recruitment. There was a significant linear relationship between ln(R/S) and the number of spawners for populations in the EF Andreafsky River (F_1,16 = 7.94; P = 0.012; R² = 33%) and the Gisasa River (F_1,14 = 20.46; P < 0.001; R² = 59%). Parameter estimates for the y-intercept and slope values were similar for populations in the EF Andreafsky [ln(α) = 0.620 (SE = 0.363) and β = 1.2 × 10⁻⁵ (SE = 0.4 × 10⁻⁵)] and Gisasa rivers [ln(α) = 0.715 (SE = 0.272) and β = 1.2 × 10⁻⁵ (SE = 0.3 × 10⁻⁵)]. The α-parameters calculated from the y-intercepts were 1.86 and 2.04, respectively, for the two populations. These data suggest that the number of spawners most likely to produce the maximum sustainable harvests (S_MSY) were approximately 24,000 (68% CI = 8,000–55,000) in the EF Andreafsky and 27,000 (68% CI = 14,000–47,000) in the Gisasa River populations (Figure 13). The sustainable escapement goal for the Chum Salmon population in the EF Andreafsky River is currently >40,000 spawners (Estensen et al. 2018), as recommended by Fleischman and Evenson (2010), which falls within our estimated S_MSY range. There is no escapement goal for the Chum Salmon population in the Gisasa River. Maximum recruitment (1/β) was predicted to occur with approximately 85,000 and 83,000 spawners for the two populations. Paired dashed curves illustrated the approximate 68% CIs resulting from parameter estimate uncertainty (Figure 13). Both populations of Chum Salmon exhibited high contrast in both number of spawners (>23) and recruits (>19 for the EF Andreafsky River population and >11 for the Gisasa River population when the outlier was censored). The tributaries exhibited similar patterns in over- and underproduction during the observable time frame, but there were no overall trends in production anomalies for either tributary (EF Andreafsky River: β_year = 0.044, 95% CI = −0.053–0.141; Gisasa River: β_year = 0.083, 95% CI = −0.025–0.191).

Chinook and Chum Salmon populations have been monitored and sampled for over two decades at the EF Andreafsky and Gisasa River weirs in the lower Yukon River. During this period the marine habitats of Yukon River salmon have experienced large-scale variability in temperature, sea ice cover, and ecology (Mueter and Litzow 2008; Royer and Grosch 2009; Stabeno et al. 2017), the annual abundance of salmon returning to the Yukon River has varied by a factor of >5 for Chinook Salmon and >8 for Chum Salmon (Pfisterer et al. 2017), annual fish harvest levels have varied in response (Evenson et al. 2009; Estensen et al. 2018), and numerous fishery regulatory changes have been enacted (Estensen et al. 2018). These data and analyses we present here suggest overall long-term resilience of these salmon populations despite their experiencing large annual variations in most metrics we examined, including escapement levels of the Chinook Salmon population in the EF Andreafsky River and both populations of Chum Salmon, run timing for all four populations (Figures 3 and 9), Chinook Salmon length at age (Figure 5), and Chum Salmon age structure (Figure 10). Significant declines over time were observed in Chinook Salmon escapement levels for the Gisasa River population, and Chinook Salmon age structure is trending significantly younger in both populations (Figure 4), which is a widespread trend for Alaska Chinook Salmon populations in general (Ohlberger et al. 2018). Production varied considerably from year to year, with the EF Andreafsky River Chinook Salmon population and both Chum Salmon populations averaging greater than one recruit per spawner during our time series, while the Gisasa River Chinook Salmon population averaged just one recruit per spawner (Figures 6 and 12). The Gisasa River Chinook Salmon population is the only one of the four salmon populations we examined here that appears to be on a declining trend.

Chinook and Chum Salmon populations occur together in the EF Andreafsky, Gisasa, and many other rivers in the Yukon River drainage (Brown et al. 2017; Larson et al. 2017); however, patterns of production and abundance among populations are distinctly different for the two species. Uncorrelated patterns of annual escapement for the two Chinook Salmon populations we examined indicate that they vary independent of each other and of the larger run in the Yukon River (Figure 2). These findings suggest that Chinook Salmon populations in the Yukon River are behaving consistent with metapopulation theory as it relates to Pacific salmon and would tend to buffer the annual abundance of drainage-wide runs against wildly oscillating highs and lows (Schtickzelle and Quinn 2007; Schindler et al. 2010). The annual abundance of specific spawning populations, however, would lack management utility for corroboration of main-stem run strength indicators. By contrast, the strongly correlated patterns of escapements between Chum Salmon populations and the larger Yukon River summer run (Figure 8), as well as the highly correlated patterns of brood year production (Figure 12), make it appear that the species is behaving like a single drainage-wide population, a phenomenon similarly discussed by Shotwell and Adkison (2004). The high level of synchrony observed among Chum Salmon populations in the Yukon River would not be expected to effectively buffer the annual abundance of drainage-wide runs, but would provide greater management utility.

These species-specific patterns of annual population abundance, however, are thought to be artifacts of life history. Juvenile Chinook Salmon occupy their natal streams or other freshwater habitats for a year prior to migrating to sea (Healey 1991). Each population would therefore be subject to unique environmental conditions influencing growth and survival during their first year (Bradford et al. 2009; Daum and Flannery 2011; Neuswanger et al. 2015). Juvenile Chinook Salmon from different populations would not experience a common environment until they migrated to sea during their second summer. Murphy et al. (2017) documented a significant relationship between the abundance of juvenile Chinook Salmon in the eastern Bering Sea and subsequent brood-year returns in the Yukon River, indicating relatively stable marine survival dynamics shortly after they get to sea. Chum Salmon, on the other hand, migrate to sea during their first summer, so each cohort within the drainage would be experiencing common environmental variables much sooner in life than Chinook Salmon. The large and synchronous variation we observed in production for the two Chum Salmon populations (Figure 12), however, suggests that survival at sea may be unique for each brood year, but common to all populations in the river. Population dynamics for these two species are clearly very different.

Annual run timing of Chinook and Chum Salmon populations in the EF Andreafsky and Gisasa rivers has varied by approximately 1 wk one way or the other around the mean but no significant trends toward earlier or later run timing were detected (Figures 3 and 9). The Gisasa River salmon runs take place an average of approximately 4 d and 5 d later than in the EF Andreafsky River for Chinook and Chum Salmon, respectively. Chinook and Chum Salmon heading to the Gisasa River are known to migrate upstream at rates averaging approximately 40 km/d (Eiler et al. 2015; Larson et al. 2017). Given that the distance between the mouths of the EF Andreafsky and Gisasa rivers is approximately 730 km, the migration would require 18 d to complete. The implication of this is that the Chinook and Chum Salmon populations migrating to the Gisasa River must begin their spawning migrations approximately 2 wk before the populations in the EF Andreafsky River.

An empirical migration model developed by Mundy and Evenson (2011) predicts that, with earlier ice retreat and warming of the eastern Bering Sea, spawning migrations of adult Chinook Salmon into the Yukon River should begin earlier in the summer than during previous colder periods. Actual spawning timing, however, is thought to be primarily constrained by selective factors other than the opportunity for adults to migrate (Cushing 1990; Quinn 2005). Stream temperature during migration and in the spawning area influences the metabolic rates of adult spawners, which limits the time available to spawn prior to depleting stored energy (Hasler et al. 2012). Stream temperature may be different from the temperature the eggs experience after they are deposited in redds within the hyporheic zone (Stanford and Ward 1988; Zimmerman and Finn 2012). The hyporheic zone temperature strongly influences the time required for incubation and eventual emergence of juveniles the next spring (Murray and McPhail 1988; Zimmerman and Finn 2012). Ultimately, the ecological conditions required for juvenile survival, a temporal match or mismatch between emergence of juveniles and the availability of their invertebrate food sources, are the selective constraints dictating the optimum spawning time for each stream (Cushing 1990; Bryant 2009; Tillotson and Quinn 2018). Spawning time, and the migration that precedes it, may eventually change for these populations of Chinook and Chum Salmon, but our data suggest it has been relatively stable in recent times.

The large-scale, size-selective commercial fishery for Chinook Salmon that took place every year in the lower Yukon River between 1961 and 1999 (Evenson et al. 2009) has generated substantial concern within the scientific community about the long-term genetic consequences for the species. Age and length at maturity are widely considered to be heritable traits in salmonid fishes in general (Carlson and Seamons 2008), and Chinook Salmon specifically (Hankin et al. 1993; Berejikian et al. 2011), and would therefore be expected to change under the influence of size- and age-selective fisheries. Modeling approaches to evaluate the long-term effects of size-selective fisheries on Chinook Salmon populations suggest that declines in age and length distributions and production capacity would occur but would probably require several decades of exploitation at rates of ≥0.5 to be detectible (Hard et al. 2008, 2009; Bromaghin et al. 2011b). When we began this retrospective analysis, we recognized that our operational time frame was not sufficient to detect fisheries-induced evolutionary changes, but we were interested in detecting the influence of the size-selective fishery in the lower Yukon River on the subsequent age structure of Chinook Salmon escapements in their spawning streams, which is the basis for expecting evolutionary change in populations. We were empirically testing the null hypothesis that there would be no trend in age structure over time versus the alternative hypothesis that age structure would trend toward larger and older fish. The conditions appeared to be ideal for this test, with Chinook Salmon exploitation rates in the lower river estimated to be 0.33 and 0.40 for the EF Andreafsky and Gisasa River populations, respectively, during the first few years of our time series, and declining to rates of 0.04 and 0.08 during the last few years. In addition, large-mesh gillnets were used to harvest Chinook Salmon during the early and middle periods of our time series and were reduced in the drainage by regulation during the later years. These conditions led us to expect a trend over time of escapements composed of more large and old fish than young and small fish. As described above and illustrated in Figure 4, we essentially observed the opposite.

These results suggest that during the operational period of these weirs, environmental factors were more strongly influencing the declining age structure of annual Chinook Salmon escapements than the opposite effects we were expecting given the progressive decline of the size-selective fishery. Siegel et al. (2017, 2018) used annual scale increments to analyze retrospective annual growth histories of Chinook Salmon from two western Alaska rivers, one being the EF Andreafsky River. Their analyses indicated that warm summers in the Bering Sea were correlated with fast growth during the first 2 y at sea, which increased the probability of maturing a year earlier than fish experiencing cooler summers at sea. Given this scenario, if the marine environment experienced by these Chinook Salmon populations was becoming warmer during the time period of weir operations, we would expect to observe a number of changes in age structure of the escapements, including 1) an increase in the proportion of age 4 fish, which we observed in both populations; 2) a decline in the proportion of age 6 and age 7 fish, which we observed in both populations although the trend was not significant for age 6 fish; and 3) the proportion of age 5 fish might be unchanged if it received enough early maturing fish that would have matured at age 6 in a cooler environment to equalize the number of fish maturing at age 4 that would have matured at age 5 in a cooler environment. The proportion of age 5 fish declined slightly in both populations but neither trend was significant. The age 6 proportions of the escapements would not be expected to equalize in the same way because the age 7 proportions were always small and would have contributed few early maturing individuals. We might also expect to see an increase in the proportion of age 4 females in escapements; however, females make up approximately 25% of the age 5 component most years (see Text S2, Supplemental Material), so the increase would probably be very subtle. All but one of the trends we observed in Chinook Salmon age structure are consistent with the findings of Siegel et al. (2017, 2018), suggesting a trend toward earlier maturation in a warming environment.

Spawner recruit analyses of Chinook Salmon appeared to be successful for the EF Andreafsky River population, with a significant relationship between ln(R/S) and the number of spawners in the initial regression, and what appeared to be adequate contrast in numbers of spawners (5.1) and recruits (7.3) to capture the traditional compensatory relationship we have come to expect with Pacific salmon populations (Figure 7; Ricker 1975; Quinn and Deriso 1999; Needle 2002). Our estimate of the number of spawners most likely to produce the maximum yield (S_MSY) was 1,687, which is similar to the value presented by Siegel et al. (2017; S_MSY = 1,701) and less than the current sustainable escapement goal (SEG) of 2,100–4,900 that was developed using the percentile method (Volk et al. 2009; Estensen et al. 2018). The 17-y time series for the Chinook Salmon population on the EF Andreafsky River revealed an average of 1.36 (SE = 0.270) recruits per spawner with no indication of a downward trend in population abundance.

Spawner recruit analyses of Chinook Salmon in the Gisasa River was quite different, though, revealing no significant relationship between ln(R/S) and the number of spawners and appeared to have inadequate contrast in the number of spawners (2.7) to reveal the underlying relationship that should be evident during years of high spawner abundance (Needle 2002). Our data, however, reveal several lines of evidence suggesting that the Chinook Salmon population in the Gisasa River has been struggling in recent years. First, the age structure is becoming younger with age 6 and 7 fish (age classes dominated by large females) declining and being proportionally replaced by age 4 fish, which are predominantly male. These changes in age class representation are causing the average annual age of the escapement to decline as well (Figure 4). Younger and smaller females on the spawning area would be expected to have a lower average fecundity (Jasper and Evenson 2006; Bromaghin et al. 2011a). Finally, given these demographic changes to the spawning population, we would expect low productivity, which these analyses confirm (Figure 6). During our 16-y time series of Chinook Salmon production data on the Gisasa River, the population averaged 1.00 (SE = 0.105) recruit/spawner, a level that would not support a fishery without declining in abundance. It is curious that both Chinook Salmon populations exhibit the same demographic trends described here, but one has maintained average production >1 recruit/spawner and the other has not.

Spawner recruit analyses of Chum Salmon populations in the EF Andreafsky and Gisasa rivers indicate great similarity among populations with a strong correlation of annual periods of high and low production (Figure 12) and statistically similar relationships between spawners and subsequent recruits when the extreme outlier from the 2001 brood year in the Gisasa River was censored (Figure 13). Data from both populations contained relatively high contrast with numbers of spawners and recruits for both populations, which is critical for capturing the underlying relationships in the model (Needle 2002). These analyses are germane to the period from 1994 to 2016. Spawner recruit analyses of the Chum Salmon population in the EF Andreafsky River were initially conducted by Clark (2001), who focused on the period from 1972 to 2000 for both forks of the Andreafsky River combined, and later using a more sophisticated analytical approach by Fleischman and Evenson (2010), who focused on the time period 1972–2007 for the EF Andreafsky River specifically. Very little of the weir-based data on escapement and age structure were available to Clark (2001) and somewhat more to Fleischman and Evenson (2010). Both previous analyses reconstructed escapement and age structure data back to 1972 from a relationship between aerial survey data and other escapement indices from the Andreafsky River and the Anvik River Chum Salmon escapement estimates, which were the only consistent quantitative data in the drainage prior to the mid-1990s when Pilot Station sonar program became functional (Pfisterer et al. 2017). Fleischman and Evenson (2010) present reconstructed escapement estimates suggesting much greater numbers of Chum Salmon in the EF Andreafsky River in the past than we are currently observing. The spawner recruit curves they calculated resulted in median S_MSY estimates at approximately 63,000 spawners, more than twice the values we calculated using only weir data. They recommended a lower-bound sustainable escapement goal of 40,000 Chum Salmon, which was adopted in 2010 (Estensen et al. 2018), but recognized the challenge of managing for this goal in the mixed-stock fishery in the lower Yukon River. Our analyses, and those of others (Shotwell and Adkison 2004; Fleischman and Evenson 2010), indicate that Chum Salmon experience highly variable production dynamics that appear to be driven much more by marine environmental conditions than by the fishery.

Long-term data sets are extraordinarily useful in fisheries research for understanding the behavior of populations over time. The weir projects on the EF Andreafsky and Gisasa rivers were initiated in 1994 (Tobin and Harper 1995; Melegari and Wiswar 1995) based on recommendations in fisheries management plans developed for the Yukon Delta (USFWS 1992) and Koyukuk (USFWS 1993) NWRs, respectively. The original goal was to provide up to 5 y of high-quality escapement and demographic data on Pacific salmon populations as a means to address one of the essential purposes for Alaska's NWRs, as specified by the Alaska National Interest Lands Conservation Act of 1980: to conserve fish and wildlife populations and habitats in their natural diversity. Despite the ambiguity of the stated purpose, the salmon populations continue to return to the rivers to spawn each year and the free-flowing rivers continue to support large populations of Chinook and Chum Salmon and other fishes. In the early years the weirs provided information on abundance, run timing, and demographic qualities at a time when few other similar data sources were available (e.g., Horne-Brine et al. 2011; Estensen et al. 2012). There was a reluctance to shut down the weir projects after 5 y of operation, in part because each year was different from one before and we did not yet understand the full range of natural variability of these salmon populations. Funding continued to be secured through various sources such that these two weirs have now been in operation monitoring the Chinook and Chum Salmon populations for 25 y, several generations of each species. These data and our analyses here establish a current benchmark detailing patterns of variation of abundance, run timing, age structure, size, and production that will allow those in the future to monitor changes that will inevitably come with time and to make decisions on resource use that conserves these habitats and the fish populations that use them.

Please note: The Journal of Fish and Wildlife Management is not responsible for the content or functionality of any supplemental material. Queries should be directed to the corresponding author for the article.

Table S1. Demographic sample data for Chinook Oncorhynchus tshawytscha and summer Chum Salmon O. keta collected between 1994 and 2016 from weirs in the East Fork Andreafsky and Gisasa rivers, tributaries of the Yukon River in Alaska.

Found at DOI: https://doi.org/10.3996/072019-JFWM-064.S1 (1.79 MB XLSX).

Table S2. Daily escapement count data for Chinook Oncorhynchus tshawytscha and summer Chum Salmon O. keta collected between 1994 and 2016 from weirs in the East Fork Andreafsky and Gisasa rivers, tributaries of the Yukon River in Alaska.

Found at DOI: https://doi.org/10.3996/072019-JFWM-064.S2 (173 KB XLSX).

Table S3. Escapement-year age structure data for Chinook Oncorhynchus tshawytscha and summer Chum Salmon O. keta collected between 1994 and 2016 from weirs in the East Fork Andreafsky and Gisasa rivers, tributaries of the Yukon River in Alaska.

Found at DOI: https://doi.org/10.3996/072019-JFWM-064.S3 (33 KB XLSX).

Table S4. Brood-year production data for Chinook Oncorhynchus tshawytscha and summer Chum Salmon O. keta collected between 1994 and 2016 from weirs in the East Fork Andreafsky and Gisasa rivers, tributaries of the Yukon River in Alaska.

Found at DOI: https://doi.org/10.3996/072019-JFWM-064.S4 (24 KB XLSX).

Text S1. Equations for analyses of stratified demographic data for Chinook Oncorhynchus tshawytscha and summer Chum Salmon O. keta samples collected between 1994 and 2016 at weirs in the East Fork Andreafsky and Gisasa rivers, tributaries of the Yukon River in Alaska.

Found at DOI: https://doi.org/10.3996/072019-JFWM-064.S5 (34 KB DOCX).

Text S2. Justification for a policy classifying all age 4 Chinook Salmon Oncorhynchus tshawytscha as males in samples collected between 1994 and 2016 at weirs in the East Fork Andreafsky and Gisasa rivers, tributaries of the Yukon River in Alaska.

Found at DOI: https://doi.org/10.3996/072019-JFWM-064.S6 (593 KB DOCX).

Reference S1.Carlson JG. 2015. Abundance and run timing of adult salmon in the Gisasa River, Koyukuk National wildlife Refuge, Alaska, 2014. Fairbanks: U.S. Fish and Wildlife Service, Alaska Fisheries Data Series Number 2015-3.

Found at DOI: https://doi.org/10.3996/072019-JFWM-064.S7 (841 KB PDF).

Reference S2.Carlson JG. 2017. Abundance and run timing of adult salmon in the Gisasa River, Koyukuk National wildlife Refuge, Alaska, 2016. Fairbanks: U.S. Fish and Wildlife Service, Alaska Fisheries Data Series Number 2017-3.

Found at DOI: https://doi.org/10.3996/072019-JFWM-064.S8 (1.37 MB PDF).

Reference S3.Flannery BG, Beacham TD, Holder RR, Kretschmer EJ, Wenburg JK. 2007. Stock structure and mixed-stock analysis of Yukon River Chum Salmon. Anchorage: U.S. Fish and Wildlife Service, Alaska Fisheries Technical Report Number 97.

Found at DOI: https://doi.org/10.3996/072019-JFWM-064.S9 (608 KB PDF).

Reference S4.Maschmann GF. 2008. Abundance and run timing of adult Pacific salmon in the East Fork Andreafsky River, Yukon Delta National Wildlife Refuge, Alaska, 2007. Fairbanks: U.S. Fish and Wildlife Service, Alaska Fisheries Data Series Number 2008-6.

Found at DOI: https://doi.org/10.3996/072019-JFWM-064.S10 (362 KB PDF).

Reference S5.McDowell S, Signorini S, Pace S, Borchardt J. 1987. Yukon Delta processes: physical oceanography. National Oceanic and Atmospheric Administration, Outer Continental Shelf Environmental Assessment Program 57:335–661.

Found at DOI: https://doi.org/10.3996/072019-JFWM-064.S11 (12.46 MB PDF).

Reference S6.Mears JD. 2015. Abundance and run timing of adult Pacific salmon in the EF Andreafsky River, Yukon Delta National Wildlife Refuge, Alaska, 2014. Fairbanks: U.S. Fish and Wildlife Service, Alaska Fisheries Data Series Number 2015-5.

Found at DOI: https://doi.org/10.3996/072019-JFWM-064.S12 (950 KB PDF).

Reference S7.Mears JD, Morella J. 2017. Abundance and run timing of adult Pacific salmon in the EF Andreafsky River, Yukon Delta National Wildlife Refuge, Alaska, 2015. Fairbanks: U.S. Fish and Wildlife Service, Alaska Fisheries Data Series Number 2017-2.

Found at DOI: https://doi.org/10.3996/072019-JFWM-064.S13 (1.26 MB PDF).

Reference S8.Melegari JL. 2008. Abundance and run timing of adult salmon in the Gisasa River, Koyukuk National Wildlife Refuge, Alaska, 2007. Fairbanks: U.S. Fish and Wildlife Service, Alaska Fisheries Data Series Number 2008-13.

Found at DOI: https://doi.org/10.3996/072019-JFWM-064.S14 (620 KB PDF).

Reference S9.Melegari JL. 2011. Abundance and run timing of adult salmon in the Gisasa River, Koyukuk National wildlife Refuge, Alaska, 2010. Fairbanks: U.S. Fish and Wildlife Service, Alaska Fisheries Data Series Number 2011-5.

Found at DOI: https://doi.org/10.3996/072019-JFWM-064.S15 (728 KB PDF).

Reference S10.Melegari ML, Wiswar DW. 1995. Abundance and run timing of adult salmon in the Gisasa River, Koyukuk National Wildlife Refuge, Alaska, 1994. Fairbanks: U.S. Fish and Wildlife Service, Alaska Fisheries Data Series Number 95-1.

Found at DOI: https://doi.org/10.3996/072019-JFWM-064.S16 (686 KB PDF).

Reference S11.Schindler D, Krueger C, Bisson P, Bradford M, Clark B, Conitz J, Howard K, Jones M, Murphy J, Myers K, Scheuerell M, Volk E, Winton J. 2013. Arctic–Yukon–Kuskokwim Chinook Salmon research action plan: evidence of decline of Chinook Salmon populations and recommendations for future research. Prepared for the AYK Sustainable Salmon Initiative, Anchorage.

Found at DOI: https://doi.org/10.3996/072019–JFWM-064.S17 (8.25 MB PDF).

Reference S12.Siegel JE. 2017. Determinants of life history variability in the Chinook Salmon (Oncorhynchus tshawytscha) of western Alaska. Master's thesis. Fairbanks: University of Alaska.

Found at DOI: https://doi.org/10.3996/072019-JFWM-064.S18 (2.75 MB PDF).

Reference S13.Tobin III JH. 1994. Construction and performance of a portable resistance board weir for counting migrating adult salmon in rivers. Kenai: U.S. Fish and Wildlife Service, Alaska Fisheries Technical Report Number 22.

Found at DOI: https://doi.org/10.3996/072019-JFWM-064.S19 (1.55 MB PDF).

Reference S14.Tobin III JH, Harper KC. 1995. Abundance and run timing of adult salmon in the EF Andreafsky River, Yukon Delta National Wildlife Refuge, Alaska, 1994. Kenai: U.S. Fish and Wildlife Service, Alaska Fisheries Progress Report Number 95-5.

Found at DOI: https://doi.org/10.3996/072019-JFWM-064.S20 (1.35 MB PDF).

Reference S15.Tobin III JH, Harper KC. 1996. Abundance and run timing of adult salmon in the EF Andreafsky River, Yukon Delta National Wildlife Refuge, Alaska, 1995. Kenai: U.S. Fish and Wildlife Service, Alaska Fisheries Progress Report Number 96-1.

Found at DOI: https://doi.org/10.3996/072019-JFWM-064.S21 (1.55 MB PDF).

Reference S16.Tobin III JH, Harper KC. 1997. Abundance and run timing of adult salmon in the EF Andreafsky River, Yukon Delta National Wildlife Refuge, Alaska, 1996. Kenai: U.S. Fish and Wildlife Service, Alaska Fisheries Progress Report Number 97-1.

Found at DOI: https://doi.org/10.3996/072019-JFWM-064.S22 (1.59 MB PDF).

Reference S17.[USFWS] U.S. Fish and Wildlife Service. 1992. Fisheries management plan, Yukon Delta National Wildlife Refuge. Kenai, Alaska: U.S. Fish and Wildlife Service.

Found at DOI: https://doi.org/10.3996/072019-JFWM-064.S23 (6.72 MB PDF).

Reference S18.[USFWS] U.S. Fish and Wildlife Service. 1993. Fisheries management plan, Koyukuk National Wildlife Refuge. Fairbanks, Alaska: U.S. Fish and Wildlife Service.

Found at DOI: https://doi.org/10.3996/072019-JFWM-064.S24 (2.96 MB PDF).

Reference S19.VanHatten GK. 2002. Abundance and run timing of adult salmon in three tributaries of the Koyukuk River, Alaska, 2001. Fairbanks: U.S. Fish and Wildlife Service, Alaska Fisheries Data Series Number 2002-5.

Found at DOI: https://doi.org/10.3996/072019-JFWM-064.S25 (361 KB PDF).

Reference S20.Wilson J. 2017. Blind Creek Chinook Salmon enumeration weir, 2016. Whitehorse, Yukon, Alaska: Yukon River Panel, Report CRE-37-16.

Found at DOI: https://doi.org/10.3996/072019-JFWM-064.S26 (1.06 MB PDF).

Reference S21.Wiswar DW. 2000. Abundance and run timing of adult salmon in the Gisasa River, Koyukuk National Wildlife Refuge, Alaska, 1999. Fairbanks: U.S. Fish and Wildlife Service, Alaska Fisheries Data Series Number 2000-1.

Found at DOI: https://doi.org/10.3996/072019-JFWM-064.S27 (682 KB PDF).

Reference S22.Zabkar LM, Harper KC. 2003. Abundance and run timing of adult Pacific salmon in the East Fork Andreafsky River, Yukon Delta National Wildlife Refuge, Alaska, 2001 and 2002. Kenai: U.S. Fish and Wildlife Service, Alaska Fisheries Data Series Number 2003-5.

Found at DOI: https://doi.org/10.3996/072019-JFWM-064.S28 (424 KB PDF).

The U.S. Fish and Wildlife Service, including the Fairbanks and Kenai Fish and Wildlife Conservation Offices and the Yukon Delta and Koyukuk National Wildlife Refuges, have provided essential personnel, funding, and administrative support for operating these weirs. Major operational funds have also been provided for many years by the Office of Subsistence Management through their Fisheries Resource Monitoring Program. Additionally, many individuals have spent their summers on the weirs counting fish and collecting the biological data essential for the types of analyses we have conducted here. RJ Henszey (USFWS) created the map figure. The Alaska Department of Fish and Game has provided essential support for these weirs as well by permitting weir operations each year and by aging all the scales that have been collected. As is almost always the case, the focus and clarity of this manuscript was improved following the constructive comments and suggestions of three anonymous reviewers and the Associate Editor. All of these contributions are greatly appreciated.

Any use of trade, product, website, or firm names in this publication is for descriptive purposes only and does not imply endorsement by the U.S. Government.