Unnatural selection of salmon life histories in a modified riverscape

Abstract Altered river flows and fragmented habitats often simplify riverine communities and favor non‐native fishes, but their influence on life‐history expression and survival is less clear. Here, we quantified the expression and ultimate success of diverse salmon emigration behaviors in an anthropogenically altered California river system. We analyzed two decades of Chinook salmon monitoring data to explore the influence of regulated flows on juvenile emigration phenology, abundance, and recruitment. We then followed seven cohorts into adulthood using otolith (ear stone) chemical archives to identify patterns in time‐ and size‐selective mortality along the migratory corridor. Suppressed winter flow cues were associated with delayed emigration timing, particularly in warm, dry years, which was also when selection against late migrants was the most extreme. Lower, less variable flows were also associated with reduced juvenile and adult production, highlighting the importance of streamflow for cohort success in these southernmost populations. While most juveniles emigrated from the natal stream as fry or smolts, the survivors were dominated by the rare few that left at intermediate sizes and times, coinciding with managed flows released before extreme summer temperatures. The consistent selection against early (small) and late (large) migrants counters prevailing ecological theory that predicts different traits to be favored under varying environmental conditions. Yet, even with this weakened portfolio, maintaining a broad distribution in migration traits still increased adult production and reduced variance. In years exhibiting large fry pulses, even marginal increases in their survival would have significantly boosted recruitment. However, management actions favoring any single phenotype could have negative evolutionary and demographic consequences, potentially reducing adaptability and population stability. To recover fish populations and support viable fisheries in a warming and increasingly unpredictable climate, coordinating flow and habitat management within and among watersheds will be critical to balance trait optimization versus diversification.

"rearing locations", then used these values to train a quadratic DFA (QDFA) assuming equal priors. Using jack-knife resampling, the QDFA predicted 83% of the "SJR rearers" correctly (80% correctly overall). We calculated equivalent metrics for the natal portions of otoliths from all adults that had been initially classified as strays from Mokelumne and Feather River Hatcheries (i.e., the individuals most likely to represent misclassified fry) then used the QDFA to predict their rearing locations. This reclassified 7.6% (23 of 304 individuals) as "SJR rearers". However, given that this is a new approach based on few "knowns" (n=6 adults that we were confident had reared in the SJR, comprising 33 individual spot measurements in putative SJR water), we only accepted the reclassification if all spots were within the range of SJR water values (Dataset S1) and if the otolith microstructure (Barnett-Johnson et al. 2007) also suggested wild origin. As a result, only eight individuals were finally reclassified as Stanislaus River returns that had dispersed as newly emerged fry (~30mm FL) and reared in the SJR.
Using otolith Sr 87 /Sr 86 to estimate size at natal and freshwater exit For returning adults, we reconstructed their size at natal exit using the otolith distance from the core to the last measurement exhibiting a Sr 87 /Sr 86 value ± SE within the range of measured values in the otoliths of known-origin juveniles or water samples from the Stanislaus River during a comparable period (see above; Dataset S1). Given that the rotary screw traps are 13.8rkm upstream of the SJR confluence (i.e. the location of the isotopic shift used to identify natal exit in the returning adults, Fig. 1), we used the last recorded "natal value" rather than interpolating among neighboring spots. Size at freshwater exit was estimated as the otolith radius when 87 Sr/ 86 Sr ratios exceeded the mean value recorded at Chipps Island (0.7077 -see Dataset 1), roughly equivalent to the fish moving into waters >0.5ppt (Hobbs et al. 2010). We used linear interpolation between neighboring spots to estimate the otolith distance at this threshold value (after Phillis et al. 2018). Otolith distance was then converted into fish fork length (FL) using the relationship shown in Fig. S2 (methods detailed in main text).

Adult recruitment
The number of adult recruits produced by each emigration cohort was estimated as estimated escapement (GrandTab; available at www.dfg.ca.gov) corrected for strays and return age distributions, plus ocean and inland harvest. We used year-matched age distributions obtained from adults on the Stanislaus River spawning grounds (CDFW unpublished annual age estimates for unmarked salmon for 2001-2015; Mesick et al. 2009Mesick et al. age estimates for 1996Mesick et al. -2000 to estimate the numbers of adult returns per emigration cohort. Annual ocean harvest rates for 3 and 4 year olds were estimated as total ocean harvest divided by the Sacramento Index (Pacific Fishery Management Council 2016). We assumed zero ocean harvest for 2-year old returns (advice of M. O'Farrell, NMFS, as 2 year olds tend to be below minimum size for the commercial fishery. Note that we also ran the analysis assuming equal harvest for all ages and it made little difference to the results). Inland harvest rates were assumed to be 5% each year (after USFWS 2017).
We used two data sources for estimating straying rates into the Stanislaus River: (1) otolith natal assignments from our seven emigration cohorts (see above), and (2) Constant Fractional Marking Program tag recovery data. For (1), as our otolith sample was randomly taken from unmarked fish only, we first removed adclipped spawners from total escapement estimates (querying raw numbers of adclipped fish per year from www.rmpc.org then expanding by the annual mean expansion factor ["estimated_number" field]) before applying our straying estimates and age frequencies (Table S2). For (2), CFM straying rates were available for 2010-2013 escapements at the time of writing , Palmer-Zwahlen et al. 2018. Trucking rates of hatchery fish (a major determinant of straying rate; Sturrock et al, 2019) were relatively similar in the years before and after this period, so we applied the first estimate (2010) to preceding escapement years (1997)(1998)(1999)(2000)(2001)(2002)(2003)(2004)(2005)(2006)(2007)(2008)(2009) and the last available estimate (2013) to proceeding years . While these recruitment estimates should be treated with caution, in years with coupled with otolith and CFM-derived recruitment estimates (n=7), the overall mean difference in recruitment was 1.5% (Fig. S3).
One should note that weir counts on the Stanislaus River (Peterson et al. 2017) suggest that net upstream passage of adults could be ~35% higher than the Grandtab escapement estimates we used to estimate recruitment (Table S2). As weir data were not available for all years of interest, we used Grandtab data to remain consistent across the whole dataset. It is likely that not all adults passing the weir successfully reach the spawning grounds, but these discrepancies warrant further investigation as they could mean that our juvenile survival probabilities are underestimated. However, we assumed that most error would come from the juvenile passage expansions (note large confidence intervals in Table S4), which we did incorporate into our survival estimates and their associated confidence intervals (Table S5). Importantly, for years with both weir and Grandtab escapement estimates (n=13; 2004-2010, 2012-17), they were highly correlated (r 2 = 0.92), suggesting that while absolute recruitment and survival estimates might be slightly underestimated, interannual patterns (e.g. the relationship between rearing flows and recruitment in Fig. 5) would not be altered by updating the underlying dataset.

Testing for differences in phenotype survival probabilities
We compared annual phenotype survival rates using Welch ANOVA and Games-Howell posthoc test as we could not transform the data to meet the assumption of homoscedasticity. Differences in survival within years were estimated by randomly resampling (n=100,000) simulated phenotype survival estimates, i.e. we randomly picked one possible number of recruits for a given phenotype and divided it by one possible number of emigrants of the same phenotype, and repeated the process 100,000 times. For each random sampling, we calculated the difference in survival among phenotypes (e.g. parr S -fry S). The distribution of these values represents the distribution of possible survival differences (equivalent to a t-statistic), with the average difference between them being the expected ("real") difference in survival and the proportion of comparisons exhibiting a difference in survival greater than zero being equivalent to a one-tailed p-value (Manly 2006) (Table S6).

Fig. S1
The proportion of fry, parr, and smolt emigrants (based on expanded rotary screw trap catches) averaged by day for all years with reliable trapping data (1996,(1998)(1999)(2000)(2001)(2002)(2003)(2004)(2005)(2006)(2007)(2008)(2009)(2010)(2011)(2012)(2013)(2014) to demonstrate the general timings that these size classes leave the Stanislaus River. Subtle among-year differences in phenology are shown in Table S3.  year old returns). For the other cohorts, we assumed straying rates when CFM data were not available (see Methods above, and Table S2) and thus these recruitment estimates were deemed less reliable, and we always used otolith-derived recruitment estimates where available.    S6. Controls on juvenile phenotype expression. The proportion of the juvenile salmon that emigrated as fry (circles), parr (triangles) or smolts (squares) is shown as a function of flow variation (A) and spawner density (B); predicted relationships based on the model described in Table 1 are shown as red, blue and green lines, respectively. Plots C to E show observed vs. predicted proportions for each phenotype and the fit (r 2 value) in the bottom-right corner. S7. Annual trends in phenotype abundance, recruitment and survival. Boxplots displaying the abundance of emigrants and adult recruits by phenotype (fry, parr or smolt size at emigration from the natal river) for the seven emigration cohorts with paired otolith reconstructions, and estimated downstream survival of each phenotype (recruits per emigrant). The box represents the interquartile range, the bold line represents the median, and the whiskers represent the 95% confidence interval.   Table S2. Data used to estimate the number of adult recruits produced by each emigration cohort (primarily represented by 2, 3, and 4 year old returns). Table continues overleaf (where footnotes include all data sources). We used Constant Fractional Marking Program (CFM) and otolith (oto) derived straying rates, estimated for all fish (by escapement year) and for unmarked spawners (by emigration cohort), respectively. For years prior to CFM we applied the earliest available stray rate, and for years after the last available report we applied the last available estimate (asterisks). As such, some caution should be applied to all CFM recruitment estimates except emigration cohorts 2008 Table S5. Juvenile salmon survival between emigration from the natal stream and adult recruitment in 1996-2014. For years with paired otolith reconstructions we estimate the contribution, abundance and survival probabilities of fry, parr and smolt emigrants. Adult 'returns' represent the unmarked escapement corrected for hatchery strays and return age (Table S2); 'recruits' represent returns plus harvest. Abundance of juvenile emigrants was estimated from expanded rotary screw trap catches (Table S4). Noise in the otolith size-fork length relationship (Fig. S2) was used to generate 95% confidence intervals around phenotype contributions to the escapement and the number of recruits (in parentheses). Confidence intervals around phenotype survival probabilities incorporated this error term along with sampling (catch) and estimation (efficiency model) error. The data used to estimate the phenotype survival rates in this table are visualized in Fig. S7. Parr survival in 2008 § was assumed to be anomalously high as a result of low catches resulting in uncertain juvenile passage estimates (n=4 caught the entire season). Table S6.
Within-year differences in phenotype survival (S) based on random resampling (n=100,000 draws) of simulated phenotype survival estimates. The ratio represents the relative difference in survival rates from these draws and the p-value represents the fraction of draws where the difference in survival was greater than zero (n.s. = not significant) Additional data table S1 (separate file; "Dataset 1.csv") Reference strontium isotope (Sr 87 /Sr 86 ) values, including otoliths from known-origin juvenile salmon (natal region of the otolith only) and water samples collected in the San Joaquin basin, freshwater Delta, and bays (Fig. 1). These data were used to estimate provenance and habitat transitions in adult salmon sampled on the Stanislaus River spawning grounds. Otolith data includes the mean, minimum, maximum, and standard deviation (SD) for all individual natal measurements for each fish.