Anthropogenic disturbance of stream ecosystems, often chronic in nature, has been studied extensively. However, when disturbance is driven by more than one resource policy over many decades, feedback between habitat evolution and biological adaptation can be disrupted and ecological function affected in unforeseen ways. We analyzed over 100 years of Chinook salmon (Oncorhynchus tshawytscha) length frequency trends associated with fisheries management and changes in available spawning substrate (habitat) linked to flow regulation in a highly altered California river. Over time, salmon lengths generally decreased, fluctuating with exploitation (ocean harvest) and hatchery production rates. Female size reduction, coupled with a degrading and coarsening channel, and perching peripheral habitat related to past mining activity, indicates available spawning substrate may be too large to support the current salmon population. Assuming a salmon can move material ~10% of her body length, length frequency data and current substrate size distribution suggests that increasing salmon sizes to historic distributions could increase available spawning habitat by as much as 13%. Alternatively, decreasing spawning substrate size could support a greater portion of the current population. To test the latter hypothesis and inform future management actions, we monitored two spawning riffles where large and small gravel was placed on top of cobble. We observed immediate spawning activity increase that was more pronounced where smaller gravel was deposited. Following a decade of habitat decline, the two sites were both replenished with medium gravel. Elevated spawning use occurred immediately at both sites, commensurate with this intermediate size, further supporting our hypotheses. Sediment coarsening and habitat disconnect below dams, combined with reduced salmon size, indicate the natural spawning process may be decoupled from available habitat below dams in the foreseeable future without continuous intervention. Actively managing salmon population demographics through modified hatchery and size-selective harvest practices and developing a coarse sediment budget with size-appropriate material for regulated anadromous rivers could produce immediate benefits for ecosystem services, including salmon populations. However, these management actions will require continued maintenance and informed socio-ecological goals to remain successful.

Changes in female salmon size (1919–2021)

Length measurements of adult Chinook Salmon returning to the LAR have not been consistently recorded. However, LAR hatchery and natural production are managed as a single population. Therefore, we applied the correlation between Chinook Salmon fecundity and adult FL (Healey and Heard 1984) to 66 years of Nimbus Fish Hatchery fecundity data (Williams 2006; CDFW unpublished data) to estimate mean annual adult female Chinook Salmon body length in the LAR (1955 to 2021). In addition, Clark (1928) provided FL data from 1,423 Chinook Salmon caught by gill-net in the Sacramento River near the LAR mouth (1919–1921). Because sex was not provided for individual fish, we used the female-to-total population FL ratio from recent LAR fish surveys (CDFW unpublished data) and multiplied this by the estimated length frequencies from Clark (1928) to estimate average female size for that period, assuming that this relationship would remain constant over the period.

Exploitation of LAR Chinook Salmon production

The exploitation rate for fall run Chinook Salmon originating from the Sacramento River and its tributaries is estimated with the Sacramento Index (SI: O’Farrell et al. 2013) that was developed by the National Marine Fisheries Service (NMFS) in response to the fall run collapse in 2008–2009. The SI is an age-integrated abundance index that is used by NMFS to forecast harvest in the commercial and recreational fishery and estimate annual exploitation rates.

Hatchery and natural LAR production

To estimate LAR hatchery Chinook salmon reproduction success, we used annual Nimbus Hatchery juvenile release records and divided this by annual Nimbus Hatchery escapement numbers to provide an annual average production to individual adult (1955-2021). Escapement (adults that “escaped” the fishery and returned to spawn) data were obtained from the GrandTab dataset managed by the California Department of Fish and Wildlife; a long-term compilation of Chinook salmon escapement estimates from Sacramento-San Joaquin basin surveys (CDFW; Azat 2022). Though different methods were used across time and space, GrandTab is the only continuous historical dataset available for California Chinook salmon escapement estimates, and the primary source used by fishery management agencies (Takata et al. 2017). Similarly, to estimate natural LAR reproduction success, we used available juvenile production estimates from the spawning grounds based on LAR Rotary Screw Trap data (1993-96; 2012-2020; CDFW unpublished) and divided this by annual estimated in-river escapement numbers to provide an annual average production per individual adult naturally reproducing in the LAR. These two measurements provide a relative index of juvenile production per adult by origin (hatchery or natural reproduction).

To demonstrate potential hatchery production influence on the spawning grounds we used data from the Central Valley Constant Fractional Marking Program (CFM).  This program was initiated in 2007 to gain better information on harvest management and hatchery release strategies. Under this program, at least 25% of all hatchery-origin juveniles are marked with an adipose fin clip and tagged with a coded wire tag. The proportion of tagged fish recovered on the spawning grounds in individual tributaries is then used to estimate the proportion of hatchery origin in-river spawners (Kormos et al. 2012; Dean and Lindley 2023).

Female salmon spawning bed mobilization capabilities

To determine how changes in female salmon size alter the population’s ability to mobilize spawning bed particle sizes over time, we used the theoretical size of the largest particle that LAR female Chinook Salmon can move (Threshold Particle size) of Riebe et al. (2014).

D_T = 115(L/600)^0.62

Where D_T = Threshold particle size and L is fish length (mm).

We used this to calculate a cumulative threshold for female length frequencies in the LAR over time.

Dam impacts on sediment size availability

We documented substrate size distribution immediately below Nimbus Dam to test whether the reach became coarser over time in absence of upstream sediment inputs. We collected surface pebble counts ~ 0.8 km downstream from Nimbus Dam, following methods of Wolman (1954) to compare substrate grain size to samples collected in 2014 to those collected in 1994. We then estimated the length frequency of female Chinook Salmon that could move these particle sizes and compared them against recent length frequencies of salmon captured at the hatchery.

To determine if perched gravel bar substrates were of different sizes than those within the active main channel, and in turn affected females' ability to build nests, we collected surface pebble counts adjacent to those described above following the same methodology. We derived two metrics from each substrate sampling: median particle size (D50), and particle size at which 84 percent of the material was smaller (D₈₄). These response variables were chosen because they represent a central measure as well as the upper tail of the cumulative distribution of particle sizes that might influence the ability of spawning salmon to construct redds (Riebe et al. 2014; Olsen et al. 2005).

Gravel size effects on spawning use

To test sediment particle size effects on salmonid spawning exclusion, we compared Chinook Salmon spawning densities before and after gravel augmentation at two LAR sites where different-sized spawning material was placed in three augmentation events (2008, 2009, 2019) (Table 1). These augmentations were designed using the Spawning Habitat Integrated Rehabilitation Approach (Wheaton et al. 2004), described and studied by Zeug et al. (2014). Surface pebble counts were performed at each site before and after gravel augmentation.

Chinook Salmon spawning data were obtained from annual Reclamation surveys (2003–2020;18 spawning seasons). Two or three aerial surveys were performed each year (early November through mid-December). Photographs (1:2400 scale) were taken during each flight and used to estimate redd numbers in discrete reaches from the Interstate 80 Bridge to Nimbus Dam (~29 km; Figure 4). Aerial photography is an effective redd survey method for large streams like the LAR where water clarity is generally adequate to identify where female salmon have disturbed the substrate during redd construction (Williams 2001). Similar to the methods of Merz and Setka (2004), Chinook Salmon redd coordinates were used in conjunction with ArcMap (ESRI, Redlands, CA) to identify the redd numbers present in each LAR gravel augmentation site each year. Raw redd counts for each site were transformed to densities (i.e. redd number divided by gravel augmentation area). In addition, to account for inter-annual variation in river-wide redd production, we scaled redd densities for each year by dividing redd density for each restoration site (i.e. number of redds • m^-2) by estimated whole-river redd count from the same season (Reclamation, unpublished data); therefore, modeled redd densities are in redd • m^-2 • total river redds^-1.

To explore how substrate size may influence utilization, we estimated the maximum movable substrate size for LAR Chinook Salmon and compared these estimates with substrate size at the two sites. The 2021 LAR Chinook Salmon spawner length data were obtained from the California Department of Fish and Wildlife (CDFW) collected from fish spawned at the Nimbus Hatchery (CDFW, unpublished data). Because river and hatchery spawners are managed as a single population, we assumed that the size of salmon entering the hatchery was similar to that of natural river spawners. We used the cumulative distribution of estimated female FL to extrapolate distribution of maximum substrate size an LAR female could move (DT). We then compared substrate D84 at the four sites to estimate the proportion of each population that would be capable of moving 95% of substrate particles at each site.

Finally, to estimate how change in female spawner size over time influences spawning habitat availability, we employed the methods of Overstreet et al. (2016) to compare D_T for the 1919 and 2021 populations against D₈₄ of pebble counts collected (2021) at the Sailor Bar East and West study sites (Figure 4) to contrast the estimated fraction of the spawning beds the two populations could move (F_M).

Statistical analysis

To determine if there were relationship trends between female FL over time, we first plotted estimated FL against time to provide a visual data representation. Because the data were non-linear and patterns were visually apparent, we used piecewise linear regression model to determine whether there were significant thresholds, or ‘breakpoints’, where the regression line slope changed (Betts et al. 2007; Muggeo 2008). We used the segmented package in R to iteratively fit a standard linear model to the data, starting from initial breakpoint estimate(s). This analysis resulted in identification of breakpoints, or years when there was a change in the FL trend over time, with associated error values (Muggeo 2008). As there were also clear visual trends in exploitation rate and hatchery production when plotted over time, we used the same approach to detect dataset breakpoints. We conducted exploratory analysis of natural production and proportion of hatchery fish datasets, but no significant breakpoints were detected.

To visually depict general data trends over time in hatchery and natural Chinook salmon production datasets, we used the Lambda (λ) smoother in JMP Graphic Builder, allowing balance between long- and short-term trends (SAS 2019). This method uses a set of third-degree polynomials spliced together such that the resulting curve is continuous and smooth at the splices (knot points). The estimation is done by minimizing an objective function that is a combination of the sum of squared errors and a penalty for curvature integrated over the curve extent (Eubank 1999; Reinsch 1967). We adjusted the smoothing spline which pieces together polynomials to optimize a criterion, balancing smoothness with goodness of fit. As the value of λ decreases (toward 10), the error term of the spline model has more weight and the fit becomes less flexible and curved, approaching a straight line.

To determine if the proportion of hatchery fish on the spawning grounds was affected by time, we performed a linear regression. To explore factors determining the relative spawning use of the two sites before and after gravel augmentation with variable rock sizes, we used a general linear model (GLM). We used a Poisson distribution and log link function because count data were over-dispersed (O’Hara and Kotze 2010). In these models, the number of redds constructed was the dependent variable. Because no coarse sediment is available from upstream, we assumed erosive forces, including reservoir water releases and spawning females, would mobilize augmented sediment and reduce habitat quality over time (Merz et al. 2006). Therefore, we used Years since Augmentation, Site-Specific Surface D50, and Total River Redds as predictor variables. Finally, we used the Site Area (m²) as an offset variable. Probit p-values, which can be challenging to represent when they are small, are expressed as ‘LogWorth’ values which is log₁₀( p-value). All statistical analyses were carried out using the JMP 15.1.0 software package with significance levels set at 0.05 (SAS Institute, Cary, NC, USA).