Calibration of the Trinity River Stream Salmonid Simulator (S3) with extension to the Klamath River, California, 2006–17

First posted June 2, 2023 For additional information, contact: Director, Western Fisheries Research CenterU.S. Geological Survey6505 NE 65th StreetSeattle, Washington 98115-5016 The Trinity River is managed in two sections: (1) the upper 64-kilometer (km) “restoration reach” downstream from Lewiston Dam and (2) the 120-km lower Trinity River downstream from the restoration reach. The Stream Salmonid Simulator (S3) has been previously constructed and calibrated for the restoration reach. In this report, we extended and parameterized S3 for the 120-km section of the lower Trinity River to the confluence with the Klamath River and then to the Pacific Ocean in northern California.S3 is a deterministic life-stage structured-population model that tracks daily growth, movement, and survival of juvenile salmon. A key theme of the model is that river discharge affects habitat availability and capacity, which in turn drives density-dependent population dynamics. To explicitly link population dynamics to habitat quality and quantity, the river environment is constructed as a one-dimensional series of linked habitat units, each of which has an associated daily timeseries of discharge, water temperature, and useable habitat area or carrying capacity. In turn, the physical characteristics of each habitat unit and the number of fish occupying each unit drive (1) survival and growth within each habitat unit and (2) movement of fish among habitat units.The physical template of the Trinity River was formed by classifying the river into 910 meso-habitat units that were designated into runs, riffles, or pools. For each habitat unit, we developed a timeseries of daily discharge, water temperature, amount of available spawning habitat, and fry and parr carrying capacity. Capacity timeseries were constructed using state-of-the-art models of spatially explicit hydrodynamics and quantitative fish habitat relationships developed for the Trinity River. These variables were then used to drive population dynamics such as egg maturation and survival, and in turn, juvenile movement, growth, and survival.We estimated key movement and survival parameters by calibrating the model to 12 years (2007–18) of weekly juvenile abundance estimates from two rotary screw traps: (1) the Pear Tree trap near the downstream end of the restoration reach and (2) the Willow Creek trap site is about 40.2 km upriver from the Trinity River’s confluence with the Klamath River. The calibration consisted of replicating historical conditions as closely as possible (for example: flow, temperature, spawner abundance, spawning location and timing, and hatchery releases), and then running the model to predict weekly abundance passing the trap location. We also evaluated four alternative model structures that included either no density-dependence, density-independent movement and survival, density-dependent survival, or density-dependent movement. Akaike information criterion model selection was used to evaluate the strength of evidence for alternative model structures to simulate the observed abundance estimates.Model selection supported the conclusion that the fully density-dependent model and density-dependent survival model was better supported by the data than the no density-dependence or density-dependent movement model. Because density-dependent movement was favored in past evaluations, we focus on the results from the fully density-dependent model. Parameter estimates from this model indicated that fry were less likely than parr to move downstream and that fry moved slower. Fry had a lower daily survival probability than parr. In contrast, hatchery fish had the highest probability of movement and the lowest daily survival probability.Fitting the model to both traps individually enabled us to independently compare the fit and performance of S3 at simulating fish abundance, timing, and growth of juvenile salmon in the upper restoration reach and lower Trinity River. We obtained a better fit to the data at the Willow Creek trap site than we obtained at the Pear Tree trap site, regardless of whether we fit the model to the abundances at the Pear Tree trap or Willow Creek trap. This better fit was surprising given that the S3 input data for the upper restoration reach required fewer assumptions than fitting to the Willow Creek trap site that is farther down river. Fitting S3 to weekly abundances at the Willow Creek trap site required making assumptions about (1) extrapolating capacity-flow relationships to unmeasured habitat units; (2) spatially allocating spawners within the lower Trinity River; and (3) approximating the abundance, timing, and size of juveniles entering from tributaries. The model provided better fit to the data at the Willow Creek trap site. In the weekly abundance estimates, in relation to the S3 simulated abundances, several migration years’ (2011, 2015–17) weekly abundance estimates appeared truncated and were near or at peak annual abundances in January, suggesting that a large fraction of juveniles was migrating as early as December at the Pear Tree trap site. Some early life dynamics may not be currently incorporated into S3. For example, the estimation of abundance at the Pear Tree trap may be biased because of size selectivity. Knowing about selectivity at the Pear Tree trap could greatly improve S3’s ability to predict weekly and peak abundances each year.