Study Area

The Lookout Point Project, located on the Middle Fork Willamette River in western Oregon, is comprised of two dams and two reservoirs that are operated by USACE as part of the Willamette Valley Project. Lookout Point Dam is a large, high-head dam (84 m tall) that stores water in Lookout Point Reservoir; at full capacity the reservoir stores 477,700 acre-feet of water. Dexter Dam is a 27 m tall re-regulating dam that is located 1.6 kilometer (km) downstream of Lookout Point Dam. The re-regulating purpose of Dexter Dam produces consistent flows in downstream reaches while allowing for fluctuating flows at Lookout Point Dam that are designed to meet peak hydropower demands within the region.

The Lookout Point Project does not allow for upstream passage of adult salmon that return to the Middle Fork Willamette River for spawning, so fish are collected in the tailrace of Dexter Dam and transported by truck to upstream reaches where they are released to spawn naturally. Progeny of transported adults emerge and move downstream, eventually entering Lookout Point Reservoir. Downstream passage through Lookout Point Reservoir and Lookout Point Dam typically requires several months so juvenile Chinook salmon rear in Lookout Point Reservoir where they experience rapid growth but are susceptible to mortality from predation and parasites (Monzyk and others, 2015; Kock and others, 2019b, 2019c; Murphy and others, 2021). Downstream passage at Lookout Point Dam typically occurs when reservoir elevations decline during fall and early winter, but injury and mortality rates can be high (>50 percent; Keefer and others, 2012, 2013; Fischer and others, 2018, 2019). The goal of the 2023 deep drawdown was to facilitate improvements to downstream passage and survival through the Lookout Point Project.

Acoustic Receiver Locations

A total of 14 acoustic telemetry receivers were deployed at five locations in the Willamette River Basin to detect tagged Chinook salmon moving downstream (fig. 1). These five locations define the downstream borders of study reaches (fig. 1). Three receivers were deployed in the forebay immediately upstream of Lookout Point Dam with the area between these receivers and the head of reservoir release location at Hampton Landing being defined as the Lookout Point Reservoir reach (reach 0). Two additional receivers were deployed in the tailrace approximately 400 m downstream of the dam with the area upstream of these receivers to the forebay receivers defined as the Lookout Point Dam reach (reach 1). The tailrace release site was located immediately upstream of these two receivers. Two receivers were mounted on the face of Dexter Dam in the forebay with the area upstream of these receivers to the tailrace defined as the Dexter Reservoir reach (reach 2). Three receivers were deployed near Eugene, Oregon, in the Middle Fork Willamette River about 1 km upstream of the convergence with the Coast Fork Willamette River with the area upstream defined as the lower Middle Fork Willamette River reach (reach 3). Four receivers were deployed in the mainstem Willamette River just upstream of Willamette Falls: two each on each side of the river and spaced approximately 500 m apart. The area upstream of the uppermost pair of these receivers was defined as the mainstem Willamette River reach (reach 4). The area between uppermost pair and the lowermost pair was defined at the Willamette Falls dual array reach (reach 5).

Environmental Conditions

Several environmental variables were of interest in relation to juvenile Chinook salmon outmigration behavior and survival including water temperature, streamflow, forebay elevation at Lookout Point Dam, discharge through regulating outlets at Lookout Point Dam, and total discharge at Dexter Dam. Most environmental data were queried and pulled from the U.S. Army Corps of Engineers DataQuery 2.0 website (USACE, 2023a; https://www.nwd-wc.usace.army.mil/dd/common/web_service/) via program R (R Core Team, 2022), including water temperature (instantaneous) and flow (hourly) near the Hampton Landing release location (U.S. Geological Survey, 2022, Lookout Point forebay elevation (instantaneous), forebay temperature string data (hourly at depths of 10, 20, 30, 40, 60, 80, and 100 ft), total outflow (hourly), and regulating outlet flow (hourly), Dexter forebay elevation (instantaneous), total outflow (hourly) and spill flow (hourly), and flow in the Willamette River at Eugene (instantaneous; USGS streamgage 14158050). Instantaneous and hourly data were summarized to daily mean values. Additionally, temperature loggers (HOBO Data Loggers, Bourne, Massachusetts, Model TidbiT MX2203) were attached to the hydrophone mounts of acoustic receivers to collect local temperature data. Temperatures used for Lookout Point tailrace were averages of temperatures recorded from the two loggers at acoustic receivers positioned in the tailrace. Lookout Point Dam regulating outlet hourly gate opening data were compiled from monthly dam operations spreadsheets provided by the U.S. Army Corps of Engineers, and data were quality checked and updated as appropriate based on comparisons of reported outflows to outflows calculated from rating tables using the regulating outlet opening data and forebay elevation.

Fish Tagging and Release

Juvenile Chinook salmon were produced for this study by Oregon State University’s Wild Fish Surrogate Program (WFSP; Cogliati and others, 2023) from adult spring Chinook salmon that were spawned at the Willamette Hatchery (operated by the Oregon Department of Fish and Wildlife) in Oakridge, Oregon. The WFSP transferred eyed eggs from the Willamette Hatchery to their facility in Corvallis, Oregon in October 2022. Juvenile Chinook salmon were reared to achieve size-at-tagging that approximated fish size from Lookout Point Reservoir during fall months in previous studies (Romer and others, 2016; Kock and others, 2019b). These prior studies show a size disparity in the fall between the two life histories targeted in this study. In September and October, fork lengths for reservoir rearing fry are typically between 150 millimeters (mm) and 200 mm (Kock and others, 2019b) while fork lengths for fall migrants are typically between 80 and 120 mm (Romer and others, 2016). Mean fork length of juvenile Chinook salmon tagged and released during the study was 134.2 mm (standard deviation [SD] = 12.2 mm) with most fish between 100 mm and 160 mm (fig. 2).

Fish tagging and release occurred from late August to late October 2023 (table 1) (methods described in Liedtke and others, 2012) to target operating conditions at Lookout Point Dam when water was only passed through regulating outlets. Juvenile Chinook salmon were surgically tagged with an acoustic transmitter (Model SS400, Advanced Telemetry Systems, Isanti, Minnesota) and a passive integrated transponder (hereafter “PIT tag”; Model APT12, Biomark, Boise, Idaho). Mean weight was 25.3 grams (g) (SD = 7.1 g), and mean tag burden (total weight of acoustic transmitter and PIT tag divided by fish weight) was 1.4 percent (SD = 0.4 percent). After tagging, fish were held at the WFSP facility overnight and then transported to release sites located near the head of Lookout Point Reservoir (hereafter “head of reservoir”) and in the tailrace of Lookout Point Dam (hereafter “tailrace”). Fish releases at the head of reservoir were conducted weekly (175–190 fish released during each release week) and, in total, 1,624 tagged Chinook salmon were released at the site (table 1). Fish releases in the tailrace were conducted four times during the study (100–102 fish released during each release week) and in total, 402 tagged Chinook salmon were released at the site (table 1)

Seventy-five additional acoustic tags from the same manufacturing lot as study tags were randomly sampled, activated, and monitored with an acoustic receiver in 10-degree Celsius (°C) river water. Monitoring continued until no tags were detected on the receiver for 24 hours. The time of failure of each tag, x_h, was recorded as difference between the activation time of the tag and the timestamp of the final detection for test tag h.

Acoustic Telemetry Data Processing and Summary Statistics

Acoustic telemetry data were processed and summarized in multiple steps. First, raw ping data were screened for false positives. This yielded a dataset consisting of single acoustic ping detections recording the timestamp, acoustic tag identifier (ID), and receiver ID of each detection. These ping-level data were summarized as detection events, where a detection event is defined as the set of sequential-in-time detections of an individual tag at any of the receivers within the array of receivers denoting a monitoring location (for example, any of the three forebay receivers) without an intervening detection at any other monitoring location. A new detection event was defined if the time differences between sequential detections exceeded 60 minutes. Detection event data consisted of the acoustic tag ID, monitoring location, first detection, last detection, and number of detections.

For each individual fish, we created “space-for-time” capture histories based on the detection event data and the study design to facilitate statistical modelling using the temporally stratified multistate (TSMS) mark-recapture framework described in Hance and others (2020, 2022). The TSMS capture histories summarize the sequential-in-space detections of individual fish as they migrated through each study reach over time. The TSMS can be used with any level of temporal stratification (for example, hourly, daily, weekly). For this study, we summarized the data using daily strata. Based on the 90-day estimated tag battery life and the release dates for all fish, we used August 30, 2023, as the first day of the study and January 31, 2024, as the final day of the study. This resulted in 155 total strata (T =155).

The capture history for each individual, y_i, was a six-element vector where each element, y_i,k, was either an integer representing the day on which the individual i was first detected (or released) at each location k or zero if the individual was not detected at that location. For individuals released at Hampton Landing, y_i,0 was the day of release with y_i,0 =1 for individuals released on August 30, 2023. Likewise for individuals released in the tailrace, y_i,2 was the day of release. In event of multiple detection events at a location, we used the timestamp of the first detection of the first detection event at each location. The TSMS framework does not allow for upstream transitions and so detection events at an upstream monitoring location that occurred after a detection at a downstream monitoring location were ignored. This only occurred within Dexter Reservoir where some individuals were detected on the tailrace receivers (at Lookout Point Dam) after having been detected on the Dexter forebay receivers. We also recorded the tag activation strata for each individual as q_i. In most cases, tags were activated 48 hours before release, but, for a small number of fish, tags were active 24 hours before releases.

Statistical Model

The TSMS framework describes a data generating process starting with an initial release of fish that survive, arrive over time, and are detected at monitoring stations that mark the transition between reaches. The TSMS allows for survival probability, travel (or residence) time, and detection probability to vary based on the time of entry to given reach. We adapted elements of models described in Hance and others (2020, 2022), but the model used for this study also contains elements novel to this study. The TSMS model for Lookout Point was defined in terms of the following parameters:

To account for non-detection and the passage time for undetected fish, the TSMS model framework makes use of two recursive intermediary parameters (Hance and others, 2020). The first recursive term defines the probability of each pair of detections with no other intervening detections. We define λ_i,j,k,s,t as the probability that individual i having entered reach j on day s survived and entered reach k having on day t with an active tag having passed undetected by any monitoring stations between j and k:

for k = j+1 and

for k > j+1.

The second recursive term, χ_i,k,s defines the probability of the final detection for individual i last detected entering reach k on day s. For fish last detected entering a reach above the downstream-most monitoring location, this term accounts for the various processes that could result in a fish not being detected at any downstream location. Those possibilities include that a fish died or experienced a delayed migration until either the tag battery failed or the end of the study. For individuals last detected at the downstream-most monitoring location, we set χ_i,K,s= 1, and define the remaining terms for 0 less than or equal to [≤] k ≤ K recursively as:

For mathematical convenience in expressing the likelihood, we define two sets of indicator variables based on the observed capture histories: m_i,j,k,s,t and l_i,k,t. We let m_i,j,k,s,t = 1 if y_i,j = s, y_i,k = t, and y_i,g= 0 for j < g < k. Otherwise, we let m_i,j,k,s,t = 0. We let l_i,j,s = 1 if y_i,j = s and y_i,k = 0 for all k > j. Otherwise, l_i,j,s = 0. Abbreviating the entire set of parameters as θ allows us to define the observed data likelihood for the capture histories as:

We included an auxiliary likelihood to better inform estimates of acoustic tag battery failure by fitting a log-logistic distribution to the observed failure times in test tags:

This allowed us to set ${\delta }_{i,t}=1-F\left(t-q_i\right)={\left[1+{\left(\raisebox{1ex}{$t$}\!\left/ \!\raisebox{-1ex}{$\nu $}\right.\right)}^{\psi }\right]}^{-1}$ in the terms above with the parameters for log median tag life, v and inverse scale, ψ, shared among the test tags and study tags.

Using this framework, we were able to investigate the effect of time-varying environmental conditions and individual covariates on the apparent survival and residence time of fish in each study reach. To model the effect of covariates on apparent survival in reach k we used a general logit-link regression form:

where β_p,k are generic regression coefficients for the slope of the effect of X_i,k,p,s, the p^th generic individual and (or) time varying covariate for individual i entering reach k on day s. We used a log-logistic distribution to model the residence time probabilities for fish in each reach. That is, for s ≤ t: where the log median of residence time in each reach, μ_k,s, was estimated as a function of environmental conditions experienced on the day of entry, s, to reach k: where γ_p,k are generic regression coefficients for the slope of the effect of Z_i,k,p,s, the p^th generic individual and (or) time varying covariate for individual i entering reach k on day s. Separate inverse-scale parameters, σ_k, were estimated for each reach but shared among individuals and strata within reaches.

We chose different covariates for each study reach. For all reaches, we used individual fish length as a covariate on apparent survival, but not on median residence time. In both models, we assumed that detection probability at monitoring locations between reaches was constant for all individuals and across strata. That is, p_i,k,t = p_k ∀i,t. Time varying covariates used were as follows:

We fit the model using the Stan probabilistic programming language (Stan Development Team, 2024). Covariates were standardized by subtracting the mean and dividing by the standard deviation. We used weakly informative priors for most parameters with the exception of the parameters for v and ψ. For v, we used a normal prior with a mean of 4.5 and standard deviation of 0.5 and for ψ we use a half-normal prior with a standard deviation of 0.5. Priors for the terms of the logit-link functions were Student’s t distributions with 7 degrees of freedom and standard deviation of 2 for intercept terms and 1 for slope terms. Normal priors were chosen for the terms of μ with mean 2 for the intercept and mean of 0 and standard deviation of 1. Half-normal priors with standard deviation of 1 were used for the inverse-scale terms of the residence time distribution. We ran 4 chains with a warmup of 1,000 iterations and a sampling of 1,000 iterations for a total of 4,000 posterior samples. Model chains were verified for convergence by checking diagnostic plots.

Detection Summary

A seasonal trend was observed for fish detections as the number of fish detected per release week was initially low but increased throughout the study. The number of fish that were detected in the study area from releases that occurred during September was very low and generally increased during October (fig. 3). Of the 1,624 Chinook salmon released at the head of reservoir a total of 335 (21 percent) were detected in the Lookout Point forebay, 155 in the Lookout Point tailrace (10 percent), 139 in the Dexter forebay (9 percent), 75 (5 percent) in Eugene and 51 (3 percent) at Willamette Falls. Passage at Lookout Point Dam occurred almost exclusively through the regulating outlets; only one tagged fish passed the dam when regulating outlets were closed and turbines were operating (fig. 3). All fish detected at Eugene and Willamette Falls were from October releases and, in particular, fish released in the last 2 weeks of the study accounted for 44 of the 51 head of reservoir fish detected at Willamette Falls (fig. 3). A similar trend was apparent in the tailrace releases where the September 14 release of 100 fish saw only 11 percent of fish detected in the Dexter forebay, 3 percent in Eugene, and 1 percent at Willamette Falls. For the remaining tailrace releases, greater than 75 percent were detected in the Dexter forebay, while the last two releases in the second half of October were responsible for most Willamette Falls detections (table 2).

Environmental Conditions

The pool elevation at Lookout Point forebay steadily lowered during the drawdown until early November, when it was held near 750 ft until early December when a series of high rainfall events led to a rapid rise in reservoir elevation and ended drawdown operations. Regulating outlet flows through Lookout Point Dam were held steady through September and October, increased in November when pool elevations were at the low point, and peaked in early December before being closed off by mid-December (fig. 3). Water was passed through the turbines during mid-December to early February when the study ended (fig. 3). Lookout Point forebay temperature string data show that the forebay went isothermic around late-August, with temperatures above 18 degrees Celsius (°C) until late September (fig. 3). Water temperatures cooled through October and November. Water temperatures in Dexter Reservoir followed a similar pattern, with temperatures above 18 °C from late-August to late-September, and then cooling temperatures thereafter. Flows at both Dexter Dam and Eugene were relatively stable through September and October and increased in November and December.

Covariate Effects

The direction and magnitude of the effect of fish length on apparent survival varied among reaches (fig. 4). For reach 0, from head of reservoir release to LOP forebay, apparent survival increased with length. At mean forebay water temperature (12.7 °C), apparent survival increased from an estimate of 0.50 (90-percent confidence interval [CI]: [0.39, 0.62]) at 110 mm to 0.83 [0.78, 0.87] at 160 mm. For reach 1, from LOP forebay to LOP tailrace, apparent survival decreased with length. At mean forebay elevation (798 ft) and mean RO outflow (1,661 cubic feet per second [ft³/s]), apparent survival decreased from an estimate of 0.41 (90-percent CI: [0.26, 0.56]) at 110 mm to 0.17 [0.08, 0.28] at 160 mm. For reach 2, from the tailrace to Dexter forebay, apparent survival increased with length. At mean tailrace water temperature (11.6 °C), apparent survival increased from an estimate of 0.78 (90-percent CI: [0.66, 0.87]) at 110 mm to 0.96 [0.94, 0.98] at 160 mm. For reach 4, from Dexter forebay to Eugene, length did not have a statistically meaningful effect on apparent survival. At mean Dexter outflow (2,786 ft³/s), apparent survival was estimated to be 0.47 (90-percent CI: [0.36, 0.59]) at 110 mm and 0.61 [0.51, 0.70] at 160 mm. For reach 5, from Eugene to Willamette Falls, apparent survival increased with length. At mean Willamette River flow (5,624 ft³/s), apparent survival increased from an estimate of 0.22 (90-percent CI: [0.11, 0.36]) at 110 mm to 0.76 [0.66, 0.85] at 160 mm. In reporting the effect of time-varying environmental covariates, we use a length of 110 mm to represent the likely effect of the 2023 drawdown operations on fall migrants and 160 mm to represent reservoir rearing spring migrants.

Apparent survival for Reach 0, from release to LOP forebay, was negatively associated with forebay water temperatures (fig. 5). Median residence time for reach 0 increased with decreasing water temperature (fig. 6). Daily mean forebay water temperatures, averaged over depths to 80 ft, were above 18 °C until September 23 and apparent survival was low for all sizes through most of September (fig. 7). Estimated apparent survival from release to the forebay at 18 °C was only 0.08 (90-percent CI: [0.07, 0.10]) at the mean fork length of 134 mm, about half of that for fish at fork lengths more typical of fall migrants (110 mm: 0.04 [0.03, 0.06]), and about twice that for fish at fork length more typical of reservoir rearing fish (160 mm: 0.17 [0.12, 0.22]). At the highest temperatures observed during our study (19.6 °C on August 31), reservoir rearing sized fish would be expected to have survival of around 0.08 [0.05, 0.11]. Temperatures moderated through October, dropping below 13.5 °C after October 26, the date of our final head of reservoir release. At 13.5 °C the apparent survival for fall migrant sized fish was estimated at 0.40 [0.30, 0.50].

Apparent survival for Reach 1, from LOP forebay to LOP tailrace, increased with decreasing reservoir elevation and with increased RO outflow (fig. 5). Median residence time moved in the opposite direction with the shortest residence times associated with lowest elevations and higher RO outflows (fig. 6). Reservoir elevations reached 755 ft by October 30 and remained below this level until December 3, when a series of atmospheric rivers led to a switch to flood control operations and a rapid rise in reservoir elevations. While at the target drawdown elevation, several smaller precipitation events led to temporary increases in RO outflows necessary to maintain the target drawdown elevation. The lowest RO outflow while at the target elevation was 1,600 ft³/s. Estimated apparent survival for fish arriving in the forebay during these conditions of minimum RO outflow and minimum elevation was 0.70 [0.61, 0.79] at 134 mm fork length, 0.80 [0.66, 0.91] at 110 mm, and 0.55 [0.45, 0.65] at 160 mm (fig. 8). Median residence time was 1.7 days [1.13, 2.32] at these same conditions (fig. 9). Doubling RO outflow to 3,200 ft³/s while holding elevations at 750 ft yielded estimates of apparent survival of 0.88 [0.73, 0.97], 0.92 [0.81, 0.99] and 0.79 [0.60, 0.95] for 134 mm, 110 mm, and 160 mm lengths respectively, while the estimate of median residence time decreased to 8 hours [2.5 hours, 18 hours]. On the other hand, at reservoir elevations of 825 ft and 0 RO outflow apparent survival was estimated to be as low as 0.05 [0.005, 0.13], 0.08 [0.01, 0.22] and 0.03 [0.002, 0.09] for 134 mm, 110 mm, and 160 mm lengths respectively, while the estimate of median residence time increased to 259 days [32.1 days, 800 days]. However, these estimates are based in part on an extrapolation for tag battery failure and residence time and should not be used for planning purposes.

Apparent survival for Reach 2, from LOP tailrace to Dexter forebay, was negatively associated with forebay water temperatures (fig. 5). Median residence time for reach 2 increased with decreasing water temperature (fig. 6). Daily mean tailrace water temperatures were above 18 °C until September 26, and apparent survival was low for all sizes through most of September (fig. 10). These results are similar to Reach 0, though apparent survival in the 4-km Reach 2 was not as low as the 20-km Reach 0 for similar water temperatures. Estimated apparent survival for Reach 2 at 18 °C was 0.51 [0.45, 0.56] for 134 mm fish, 0.29 [0.21, 0.37] for 110 mm fish, and 0.74 [0.64, 0.83] for 160 mm fish. By the time the drawdown operations reached their minimum elevation, water temperatures dropped to 11 °C or below. At 10 °C the apparent survival for fall migrant sized fish was estimated at 0.85 [0.75, 0.93] and 0.98 [0.96, 0.99] for reservoir rearing sized fish.

The covariates of Dexter outflow and Willamette River flow did not result in a statistically detectable effect on apparent survival for Reach 3 and 4 (fig. 5). However, these covariates did decrease residence time in these reaches (fig. 6). The mean estimate of apparent survival through Reach 3 (Dexter Dam to Eugene) during the month of November when the drawdown was at the target elevation was 0.57 [0.42, 0.70] for 134 mm fish, 0.50 [0.33, 0.67] for 110 mm fish, and 0.63 [0.46, 0.78] for 160 mm fish. For the same period in Reach 4, the mean estimate of survival was 0.44 [0.24, 0.66] for 134 mm fish, 0.21 [0.07, 0.43] for 110 mm fish, and 0.71 [0.48, 0.89] for 160 mm fish.