Evaluation of survival estimates generated from tracking downstream migrating juvenile sea lamprey (Petromyzon marinus) with a miniature acoustic telemetry tag

Fifty-six transformed and actively outmigrating juvenile sea lampreys were collected on 22 and 27 October 2020, with drift nets deployed in Furlong Creek (Mackinac County, MI), a tributary to Lake Michigan, and transported to the U.S. Geological Survey Hammond Bay Biological Station near Millersburg, MI. Each lamprey was weighed (g), measured (total length, mm), and surgically implanted with an ELAT transmitter (2 × 12 mm, 0.08 g) custom-fabricated at the Pacific Northwest National Laboratory. The ELAT emits pings every 5 s at a carrier frequency of 416.7 kHz with an estimated battery life of 30 days (Deng et al. 2021). The 5 s ping interval was selected to maximize the probability that a transmitter was detected on any single receiver. Surgical procedures were done following the protocol recommended by Haas et al. (2023). Briefly, each lamprey was anesthetized with AQUIS-20E (10% eugenol, 0.8 mL L⁻¹ solution), and a disinfected transmitter was inserted posteriorly through a 3 mm lateral incision until completely enveloped in the lamprey's body cavity. No suturing or other method of wound closure was used following previous recommendations (Dawson et al. 2015; Moser et al. 2017). Each lamprey was revived in an oxygenated recovery tank until active swimming or suction attachment to a tank wall was observed, and then moved into holding tanks for a three-day observation period to monitor for evidence of surgical injury or tag loss (no instances occurred). Surgeries took place in 2020 on three dates: 28 November (n = 16), 1 December (n = 20), and 4 December (n = 20). After the holding period, sea lamprey were transported to the field site and released into the White River after a brief (10–20 min) acclimation period during which river water was gradually added to the holding tanks to acclimatize the lamprey to the river temperature (release dates: 1 Dec (n = 16), 4 Dec (n = 20), and 7 Dec (n = 20)). The difference in water temperature between the holding tanks and the river was typically 1–2 °C. The mean total length (±1 SD) of all lampreys that were implanted with an ELAT was 173.84 mm ± 10.48 (range = 160–205 mm), with a mean mass (±1 SD) of 7.65 g ± 1.35 (range = 5.1–12.5 g). Mean (±1 SE) tag burden (%, by mass) was 1.04 ± 0.19 (range = 0.60%–1.57%). All animal-use procedures were approved by the Michigan State University Institutional Animal Care and Use Committee via protocol #PROTO201900108.

The White River flows approximately 134 km from headwaters in Newaygo County (Michigan, USA) before emptying into Lake Michigan, with an average discharge of 12.7 m³ s⁻¹. The study site includes the lower 42 km of the river system, inclusive of three major habitat types (Fig. 1). The upper third constitutes a shallow (0.5–1.5 m deep) sandy meandering river system with a mixed-deciduous forest riparian zone interspersed with human dwellings and agricultural lands. The middle third represents a transitional environment where the river moves through a landscape dominated by wooded wetlands, emergent vegetation beds, and open water marshes. The lower third encompasses White Lake, a 1040 ha drowned-rivermouth lake that is connected to Lake Michigan via a 0.54 km hardened channel. From late April through early June, migrating sea lampreys enter the White River system from Lake Michigan and navigate to upstream spawning grounds below an impassable dam located at Hesperia, MI, approximately 60 km upstream of the river mouth.

Sea lamprey movements were detected on eight acoustic telemetry receivers (ATS model SR3001 Trident) distributed across six locations throughout the study site (Fig. 1). Beginning 1 km below the release site, three receivers (R1–R3) were placed in the meandering river section, spaced approximately 5 km apart. One receiver was placed in the riverine–wetland complex near Hilts Landing Park in Muskegon County (R4). Receivers 1–4 were attached to a tree with vinyl-coated cable and anchored in relatively deep sections of the channel (>2 m) with clear line-of-sight for 75+ m in both downstream and upstream directions. Two receiver “gates” (paired receivers with overlapping detection radii) were installed to detect movements among the major habitat types. One gate (R5 and R6, 58 m apart) was placed at the transition between White Lake and the riverine–wetland complex to detect movement into White Lake; the other gate (R7 and R8, 106 m apart) was placed at the upstream end of the shipping channel to detect movement into Lake Michigan. Two HOBO Pendant® MX temperature/light data loggers were affixed to the mooring of the upstream-most receiver to gather temperature (°C, ±0.5) and light (Lux, ±10%) data throughout the experiment. Detection data were downloaded from the receivers every two weeks during the experiment. Receivers were retrieved 40 days after the final release (16 January 2021), 10 days past the expected maximum battery life of the transmitter.

Range testing was performed at the three riverine receiver locations (R1–R3) before and during the experiment with an ELAT configured to a ping interval of 1 s. The test tag was secured to a lightly anchored rope and towed downstream behind a drifting crewed kayak for 600 m, beginning 300 m upstream of the receiver's location. GPS coordinates were recorded at the rear of the kayak (∼1.5 m downstream of the tag) at 1 s intervals using a Garmin (GPSMAP® 64x) handheld GPS unit (GCS WGS 1984), time-synched with the receivers. Detection data were obtained using JSATS Autonomous Receiver Data Filtering Software v. 1.04 (Deng et al. 2017), and the time of detection was matched to the recorded GPS location at that time. Detection range was calculated as the great circle distance (Haversine formula via R package “geosphere” v. 1.5–10; Hijmans et al. 2019) from the first (upstream range estimate) and last (downstream range estimate) transmitter detection for each drift. Two drifts were completed for each receiver, for a total of 12 records (3 receivers × 2 tows × 2 estimates per tow). Range testing was attempted but unsuccessful at lower receivers due to severe weather and restrictions related to Covid-19.

Ground speeds (km h⁻¹) were estimated for lamprey that were detected on two consecutive receivers on a single night by dividing the river distance between the receivers with the observed passage time, defined as the time difference between the last detection on the upstream receiver and the first detection on the downstream receiver. River distance (km) was determined with linear referencing software (ArcMap v. 10.7.1). To evaluate whether ground speeds differed among habitat types (Prediction 1), the mean ground speeds between receiver pairs were compared with one-way ANOVA and post-hoc Tukey's HSD tests. The ground speed estimates generated from the passage between receivers R1 and R2 were also used to back-calculate the approximate time of movement onset. Each animal's ground speed from R1 to R2 was divided by the river distance from the release point to R1 (0.98 km) to estimate the time taken to travel from the release site to R1. That time was then subtracted from the time of first detection at R1 to generate an estimated time of movement onset. To test whether daily movement onset was centered on nautical twilight (Prediction 2), we converted the 24 h circular time distribution to a 360° circle with the time of nautical twilight on the date of first detection set at 180°. We then applied a Rayleigh Test of Uniformity (R package “circular” v. 0.4–93; Agostinelli and Lund 2017) to determine if the distribution of movement onset times was significantly clustered at 180°. To avoid any bias related to the effects of transport, handling, and release on the lamprey's movement decision, movements within 4 h of release were censored from the analysis (two lampreys).

To examine the influence of river flow on movement tendency (Prediction 3), we fit a mixed-effects logistic regression (R-package “lme4”, version 1.1–33; Bates et al. 2015) relating dates when movement occurred to measured discharge for the period encompassing the first detection of a lamprey on R1 through the final detection on any receiver. We defined a movement day as the 24 h period commencing at 18:00 EST and extending overnight until 17:59 EST the following day. Each movement day during a lamprey's observation period was given a binary score (dependent variable) denoting detection (migrating = 1) or absence of detection (not migrating = 0). Discharge data were retrieved from U.S. Geological Survey discharge gage #04122200, located between R1 and R2. Predictors included mean daily discharge (Z-transformed) and the change in discharge in the 24 h prior to the start of each movement day as interactive effects, with the additive random effect of individual. Detections between release and 18:00 EST the following day were censored, as lampreys likely would not be responding to changes in the river flow when deciding whether to initiate downstream movement immediately after release into the river. This flow-effects model was compared via ΔAIC (Akaike information criterion) to a null model (a fixed-effects logistic regression with the same covariates).

We modeled the survival and transition probabilities throughout the study area with the Bayesian discrete state-space implementation of the CJS model described in Kéry and Schaub (2012). The CJS model used detection histories of all tagged sea lamprey to estimate survival probability (ϕ) between receiver locations, as well as detection probabilities (p) at each receiver location. The elements of the observation matrix with dimensions I × T [I = total number of tagged individuals (i = 1,…, I); T = number of spatial locations (t = 1,…,T)] consisted of 1 or 0 s to denote the detection (1) or lack of a detection (0) of a lamprey at a given location. Thus, p_t refers to the detection probability at the tth receiver location, whereas ϕ_t refers to the survival probability between the tth receiver location and the next downstream receiver location (t + 1). The release site was included as a spatial location for the model, and the model assumed all sea lamprey were alive at the start of the observation period. Both survival and detection probabilities were allowed to vary spatially. All tagged sea lamprey were assumed to have equal detection probabilities at each receiver. The state process for the model z_i,t (i.e., latent true state for each tagged lamprey (i) at the tth receiver) was set equal to 1 for the release location and each detection (assumed alive), but for all other locations, it was modeled following Kéry and Schaub (2012) as

The observational process conditional on the state process was defined as

For this analysis, the two receivers with overlapping detection radii at the upstream entrance to White Lake (R5 and R6) were treated as a single location. The survival probability from receivers R7 and R8 that bracket the channel to Lake Michigan (ϕ₇) was set equal to 1, given they were 30 m apart. This allowed the estimation of p₇ separately from ϕ₇. The CJS model was estimated in JAGS executed from within R (R-package “jagsUI”, version 4.0.4; Kellner 2019). The following vague prior probability distributions were specified for model parameters: ϕ ∼Uniform (0,1), p ∼Uniform (0,1). Three parallel Markov chain–Monte Carlo (MCMC) chains, each consisting of 20 000 iterations, were run from random initialization values. The first 5000 iterations were discarded as burn-in, and every 10th iteration was retained, resulting in 4500 saved samples across the chains. Chain convergence for parameters was determined by examining trace plots and scale reduction factors (R-package “coda”; version 0.19–4; Plummer et al. 2006). Maximum a posteriori probability estimates were calculated and used as parameter point estimates (R-package “bayestesR”; version 0.13–1; Makowski et al. 2019). Uncertainty in the parameter estimates was described using 95% highest posterior density intervals (R-package “coda”; version 0.19–4; Plummer et al. 2006).

We conducted simulations to evaluate model sensitivity with respect to the number of tagged individuals, the number of locations where tagged sea lampreys were released (one or two), and the magnitudes of the survival and detection probabilities. Detection data were simulated using methodologies described and code presented in Kéry and Schaub (2012). For the tagging-level evaluations, we simulated detection data for 25, 50, 100, 200, 300, 400, 500, or 1000 sea lamprey using the survival and detection probabilities (maximum a posteriori probability estimates) estimated from the CJS model fit to the White River detection data. Regarding locations where tagged sea lamprey were released, our intent was to determine whether there were benefits to changing how sea lamprey were released in the study site, owing to the limited battery life of the ELAT and uncertainty regarding the duration of juvenile sea lamprey migrations. As described above, we released all tagged sea lamprey at a single upstream site; however, if migration in some individuals was protracted, this could result in a few individuals moving into the lower section of the river prior to the transmitter battery being drained, which could affect the accuracy and precision of survival estimates in the lower reaches of the study area. An alternative release strategy would be to release lamprey at two or more locations in the river (staggered release). To explore the effect of this staggered release, we conducted simulations where we assumed half of the tagged lamprey were released at the original experimental release point (upstream of R1) and half were released at a point in the river just upstream of the entrance to White Lake (i.e., upstream of R5 and R6). The staggered release protocol was evaluated across the different tagging levels used for the sample size evaluations. As before, we used the survival and detection probabilities (maximum a posteriori probability estimates) estimated from the CJS model fit to the sea lamprey detection data to simulate the detection data. For assessing CJS model sensitivity to changes in the magnitudes of the detection and survival probabilities, detection data assuming 56 tagged sea lamprey were generated as described above, but assuming low and high survival and detection probabilities. Low parameter values were randomly drawn from uniform distributions with lower and upper limits of 0.1 and 0.3, whereas high parameter values were drawn from uniform distributions with lower and upper limits of 0.7 and 0.9.

One thousand simulations were performed for each evaluated tagging level, locations where tagged sea lampreys were released, and magnitudes of the survival and detection probabilities. The same CJS model as described previously was applied to each generated detection dataset. MCMC chain length, burn-in duration, and thinning rate were also kept the same. Chain convergence for parameter estimation was assessed only through scale reduction factors because of the total number of estimated models. Absolute error (|estimated value − true value|) was used to assess the sensitivity of the model to evaluated factors.

The observed time to stage-IV anesthesia (mean ± 1 SD) was 679 ± 175 s (range = 300–1260 s). Lampreys recovered from anesthesia after (mean ± 1 SD) 1357 ± 588 s (range = 360–3180 s). Implantation took (mean ± 1 SD) 85 ± 40 s (range = 60–240 s). There were no mortalities, apparent injuries, or abnormal behavior observed during the 3-day post-surgery holding period. Additionally, no lamprey died or exhibited behavior indicative of stress during transport from the holding facility to the field site. Upon release into the White River, sea lampreys moved quickly to the bottom and sought cover among rocks and leaf litter or burrowed into the sediment. Most sea lampreys moved quickly downstream through the first three receivers, after which several animals stopped for periods of 1–28 days (Fig. 2), with 23 lampreys stopping once, 10 stopping twice, and a single sea lamprey stopping three times (mean duration ± 1 SE, first stop, 7.47 ± 1.22 days; second stop, 7.24 ± 1.61 days). The mean duration (±1 SD) from first to last detection for any given lamprey was 6.78 ± 7.81 days. A single lamprey was detected 28 days after release. Interestingly, this animal moved through receivers R1 through R4 over the first 28 h after release and was later redetected on R4 (day 27) and R3 (day 28), appearing to move upstream. It is possible this transmitter was in the gut of a predator. Nineteen lampreys were last detected within one day of release; however, 14 of these traveled downstream to the second receiver, with one navigating to the fourth receiver within 15 h of release.

Fifty-four of 56 animals (96%) were detected on at least one receiver during the study period, with 53 first detected within 8 h after release (mean ± 1 SE: 5.64 ± 2.38 h, range: 1.36–131.66 h). The number of unique sea lamprey detected progressively decreased downstream through the study area (see inset table in Fig. 1), with only a single lamprey detected on the terminal receiver in the shipping channel connecting White Lake to Lake Michigan. There were three known missed detections on receivers, defined as a transmitter not detected on a given receiver but subsequently detected on a downstream receiver. There were no events where a transmitter failed to be detected on two consecutive receivers before reacquisition. Tag tows used to assess receiver performance for R1–R3 revealed an overall mean (±1 SE) detection radius of 31.97 ± 6.04 m (all receivers combined, upstream-looking and downstream-looking distances; Fig. 3). The mean detection radii during tag drift tests for the receivers were 37.72 ± 14.98 m (R1), 16.70 ± 3.87 m (R2), and 41.46 ± 6.71 m (R3).

The mean (±1 SE) ground speed of lampreys actively moving downstream was 1.88 ± 0.08 km h⁻¹. Consistent with Prediction 1, mean ground speeds differed significantly among receiver location pairings (ANOVA, p < 0.001, F = 58.34; Fig. 4). The fastest moving sea lamprey was observed in the riverine section between receivers R2 and R3 moving at 3.08 km h⁻¹. The mean ground speed in that section was 2.47 ± 0.10 km h⁻¹. The slowest ground speeds were observed in the river–wetland complex between R4 and R5/6 (0.86 ± 0.19 km h⁻¹). A post-hoc Tukey's HSD test indicated that differences in observed ground speed between each receiver pair were statistically significant (p < 0.001), with one exception. The ground speed of lampreys moving from R3 to R4 was not significantly different from the ground speed of lampreys moving from R4 to R5/6 (p = 0.52). Only two lampreys were detected exiting White Lake prior to the projected time of battery expiration and were censored from the above analyses. Those lampreys took 1 and 10 movement days to transit White Lake.

Thirty-eight of 56 lamprey were detected on receivers R1 and R2 on their first night of movement, allowing for estimation of the clock time of movement onset relative to nautical twilight (Prediction 2). The mean (±1 SE) onset of movement was within 16.21 ± 3.59 min of nautical twilight. This corresponded to an average latency of 58.46 ± 4.08 min after the light sensor reached a reading of 0 lx (Fig. 5). A Rayleigh test confirmed that the distribution of movement timing was significantly centered on the time of nautical twilight (±1 min, r = 0.34, p < 0.001, specified mean set to 180° representing the times of nautical twilight on the dates of observation).

The dates of movement were significantly related to higher daily mean discharge (Fig. 6; mixed effects logistic regression, Z = −3.21, p < 0.01), but not related to the difference in discharge over the previous 24 h (Z = 3.31, p = 0.19) or the interaction between the two (Z = −1.64, p = 0.10). The theoretical pseudo-R² was 0.33, with a standard deviation of 1.13 for the random effect of the individual. Additionally, this model performed better than the null (ΔAIC = 26.3).

The CJS model successfully converged, with all model parameters having potential scale reduction factors <1.01 and effective sample sizes greater than 2000, with most being greater than 4000, which we deemed sufficiently large for characterizing the uncertainty of the estimates. The posterior probability distributions for the detection probabilities for most receivers were heavily left-skewed, with the exception of the detection probability at receiver location 7 (Supplemental Material, Fig. S1). The maximum a posteriori probability estimates for the detection probabilities for the first six receiver locations were all greater than 92% [p₁ = 98% (95% HPDI (highest posterior density interval): 91%–100%); p₂ = 100% (93%–100%); p₃ = 92% (79%–99%); p₄ = 99% (79%–100%); p₅ = 97% (46%–100%); p₆ = 97% (37%–100%)]. For receiver location 7 (p₁), the estimated detection probability was 48% (8%–89%). Survival estimates were highest (>70%) in the upper part of the river and declined considerably in the river–wetland zone and drowned rivermouth lake (Fig. 7). The maximum a posteriori probability estimates of survival were ϕ₁ = 97% (95% HPDI: 90%–100%), ϕ₂ = 91% (81%–97%), ϕ₃ = 79% (66%–91%), ϕ₄ = 75% (57%–90%); ϕ₅ = 47% (26%–86%), ϕ₆ = 13% (2%–37%). While differences in survival among the reaches were large, when considered on a per-km basis, survival rates were more similar across the reaches [ϕ₁/km = 97% (95% HPDI: 89–100%), ϕ₂/km = 98% (95%–99%), ϕ₃/km = 96% (93%–98%), ϕ₄/km = 92% (95%–99%); ϕ₅/km = 86% (77%–98%), ϕ₆/km = 83% (71%–92%)]. In terms of cumulative survival for a complete transit through the study area, estimated survival was 3% (95% HPDI: 0%–9%).

All estimated models in the simulations evaluating tagging level and release locations converged on stationary and stable distributions based on scale reduction factors, with all values being less than 1.01. For the single release location, survival estimation error increased with distance from the release site, regardless of tagging level. Even at the lowest tagging level, the mean absolute error in the survival estimate was less than 5% for ϕ₁ and ϕ₂ (Supplemental Material, Fig. S2). For ϕ₅, however, the mean absolute error was greater than 10% for tagging levels of 50 lamprey and fewer. Ranges in the absolute errors (i.e., differences between maximum and minimum absolute errors within an evaluated scenario) also differed as distance from the release site increased. For ϕ₁ and ϕ₂, the range in absolute errors was generally less than 20% regardless of the tagging level, but for ϕ₅ and ϕ₆, the ranges were in excess of 50% and were as high as 80% at the lowest tagging level at ϕ₆. With respect to tagging levels, changes in mean absolute error and the range of absolute error were most noticeable as tagging levels increased from 25 tagged sea lamprey to 200–300 tagged sea lamprey (Supplemental Material, Fig. S2). Improvements in survival mean absolute error for every additional tagged lamprey were on the order of 11–16 times greater as the tagging level increased from 25 to 200 lamprey than for an increase in tagging from 200 to 1000 lamprey. For ranges of absolute errors, improvements for each additional tagged lamprey were on the order of five times greater as tagging level increased from 25 to 200 lamprey than for an increase in tagging from 200 to 1000 lamprey.

For the models evaluating a staggered sea lamprey release, mean absolute error and range of absolute errors decreased for survivals lower in the river but increased in the upstream sections. Differences in mean absolute errors between the single and staggered releases were as large as ±4% at the lowest tagging levels and were ±1% at tagging levels of 200 sea lamprey or more. For ranges of absolute errors, at the upstream locations, ranges increased by as much as 6%–12% for ϕ₁ to ϕ₄ under a staggered release strategy but decreased by as much as 28%–61% for ϕ₅ to ϕ₆, suggesting that the precision of survival estimates at these downstream locations could be substantially improved by changing to a staggered release strategy. As with the single release strategy, the largest improvements in mean absolute error and range of absolute errors for the staggered release strategy were for an increase in tagging from 25 lamprey to 200 to 300 lamprey. At larger tagging levels, relative changes in mean absolute error and range of absolute errors were smaller. Improvements in survival mean absolute error for every additional tagging level were 4–14 times greater as the tagging level increased from 25 to 200 lamprey than for an increase in tagging from 200 to 1000 lamprey. For ranges of absolute errors, improvements for every additional tagged lamprey were as large as six times greater as the tagging level increased from 25 to 200 lamprey than for an increase in tagging from 200 to 1000 lamprey.

For the evaluation of survival and detection probability magnitudes, all estimated models converged on stationary and stable distributions based on scale reduction factors, except for 0.89% of the models that fit under the low ϕ and low p scenario. These results were discarded from the saved results prior to summarization. Under the high ϕ and high p scenario, mean absolute errors in survival estimates ranged from 5% to 9%, whereas the absolute errors ranged from 26% to 48%. As with the empirical data, estimation errors increased with distance from the release location. For the high ϕ and low p scenario, mean absolute errors in survival ranged from 12% to 37%, whereas the absolute errors ranged from 52% to 82%. Conversely, for the low ϕ and high p scenario, mean absolute errors in survival ranged from 5% to 22%, and the absolute errors ranged from 26% to 85%. For the low ϕ and low p scenario, mean absolute errors in survival ranged from 12% to 27%, and the absolute errors ranged from 27% to 83%.

Consistent with prior reports for downstream migrating juvenile sea lamprey (Sotola et al. 2018; Miehls et al. 2019) and river lamprey (Lampetra fluviatilis, Pavlov et al. 2017), all observed downstream movements were nocturnal. The onset of nightly movement was clearly centered in the nautical twilight period, commencing approximately one hour after full darkness. Landward movement during the spawning migration in the Great Lakes also begins at nautical twilight, observed from sea lampreys migrating into coastal waters from offshore (Meckley et al. 2017), and during the search for suitable rivers while navigating along coastlines (Vrieze et al. 2011; Meckley et al. 2014). Similar movement timing has been reported for seaward migrating anadromous rivers and sea lampreys from captures in the Rhine River, Germany (Baer et al. 2018). Although sea lampreys sense light via paired eyes and photoreceptors distributed along the epidermis, it is likely that nightly movement rhythmicity is regulated by the pineal gland (Tamotsu and Morita 1986; Binder and McDonald 2007; Fain 2020). Delaying movement until nautical twilight likely improves avoidance of piscine and avian predators that exhibit heightened hunting activity during crepuscular periods in shallow inshore habitats (Ibbotson et al. 2006; Hammerschlag et al. 2017).

Nearly all sea lamprey initiated downstream movement on the first night after release, after which nightly movements became less frequent, suggesting lakeward migration is an intermittent process typified by frequent stopping. It is unclear if the high rate of movement on the first night was related to transport and handling. The animals used in the study were captured actively migrating downstream; consequently, their motivation to recommence migration after a holding period may have been substantial. After the day of release, lakeward movement was more likely on nights with higher river discharge. Migrating juvenile lampreys are generally more likely to migrate during fall and spring periods marked by frequent floods and to initiate movement at the start of flood events (Applegate 1950; Applegate and Brynildson 1952; Potter 1980; Hanson and Swink 1989; Goodman et al. 2015). Stopping was more frequent and of longer duration in the larger and deeper habitat zones (river–wetland complex and drowned rivermouth lake). Observed ground speeds were greatest for sea lamprey transiting the single-thread river portion of the study area, slowing as they entered the lower gradient wetland–river complex. These patterns are consistent with the hypothesis offered by Applegate and Brynildson (1952) and Miehls et al. (2019) that juvenile sea lamprey often drift during outmigration. Downstream migrating sea lamprey appear to limit active steering movements to those necessary to maintain position in the center of the stream and avoid entrainment in low-speed waters near the shore (Bracken and Lucas 2013; Johnson and Miehls 2014; Sotola et al. 2018). This strategy would allow the migrants to conserve energy to avoid predation risk during transit. Johnson et al. (2019) report increased downstream movement speed in juveniles when exposed to conspecific alarm cues in the lab, indicating active swimming to pass quickly through the risky area. Minimizing such expenditures by drifting in the absence of perceived risk and drifting on nights of high discharge would also preserve energy for active hunting once parasitic feeding commences in coastal waters.

Delayed mortality related to surgical injury, handling, or tag burden can result in underestimated survival in field studies. Implementation of the recommendations of Haas et al. (2023) greatly improved post-surgical outcomes from the laboratory study they report, most particularly the use of larger subjects. Here, no mortality or complications occurred during the post-surgical holding period, and the lampreys quickly initiated normal nocturnal downstream movement, suggesting short-term tagging effects were successfully minimized.

The probability that a transmitter is detected by a receiver is a critical component of determining survival estimates from telemetry data (Melynchuk 2012; Kessel et al. 2014; Lees et al. 2021). Unfortunately, restrictions related to the Covid-19 pandemic prevented us from evaluating receiver detection range in the river–wetland and drowned rivermouth zones. In the riverine zone, the detection radii for the receivers were low (2550 m) compared to the equipment's maximum capacity, but they encompassed the full width of the river channel. This was expected, as varying depth and a meandering channel constrain the line of sight in small river systems. Across the duration of a study, detection radii for acoustic telemetry receivers can vary in response to environmental factors in rivers (e.g., water current, substrate type, precipitation, wind) and changing water depth, which alter the maximum line of sight (Finstad et al. 2005; Chittenden et al. 2008; Gjelland and Hedger 2013; Liedtke et al. 2019; Deng et al. 2021). Despite these challenges, the modeled detection probabilities exceeded 90% in the riverine zone, declining for the receiver gates above and below White Lake. There were only three known missed detections across two locations, with 98% of all released lampreys detected on at least one receiver. These values compare favorably to prior studies of lakeward movement in juvenile lampreys using similarly sized passive integrated transponder (PIT) tags. In smaller river channels, PIT detection rates for juvenile sea lamprey ranged from 5% to 14% (Dawson et al. 2015; Miehls et al. 2019).

Disentangling spatial and temporal variation in rates of survival is a central goal of mark-recapture analysis (Lebreton et al. 1992). In the present study, modeled survival was highest in the riverine reach, appearing to decline through the river–wetland complex before falling precipitously in the drowned rivermouth lake. If the general pattern of decreasing survival in estuarine-like environs is true, it is consistent with that observed for the well-studied seaward migration of Pacific salmon, where survival rates tend to be low (0.03–0.30), decrease with greater distance traveled, and are lowest in habitats occupied by high densities of predators, including large lowland rivers and estuaries (Anderson et al. 2005; Welch et al. 2009; Goetz et al. 2015; Michel et al. 2015; Furey et al. 2016; Clark et al. 2017). It is reasonable to anticipate that juvenile sea lampreys are exposed to greater predation risk when transitioning to drowned rivermouth lakes (Great Lakes) or estuaries (Atlantic coastlines). Whether due to reversing semidiurnal tides (Atlantic) or slowed water velocity (Great Lakes), each is a region of greater retention for waterborne organisms and materials that fuel high rates of productivity and support larger and more numerous predators (Larson et al. 2013; Crump et al. 2022). The tendency to stop in these regions and await increased flow to continue downstream migration is consistent with a strategy of rapid transit through high-risk regions.

We cannot rule out the possibility that stopping in the river–wetland complex or the drowned rivermouth lake may be associated with the onset of parasitic feeding by tagged sea lamprey. Reports indicate some juvenile sea lamprey will temporarily or permanently cease downstream migration prior to reaching the terminal water body and commence parasitic feeding in small inland lakes or estuaries (Silva et al. 2013; Johnson et al. 2016; Cobo et al. 2017; King and O'Gorman 2018). As sea lampreys migrate during colder months, their potential hosts may be easier to locate and attack when swim speeds and defensive behaviors are reduced due to low water temperatures. Interestingly, the presence of these habitat types in the natal watershed could conceivably lead to increased survival during the critical first feeding period for sea lamprey. It should be noted that observations of parasitic feeding during this period are limited and merit further attention (Evans et al. 2021).

Although derived from a relatively small sample size and a limited number of receivers, we believe the survival estimates for the upper riverine zones (ϕ₁–ϕ₃) are plausible estimates of true survival for this system. We conclude this based on the observations that passage through these zones was timely (i.e., battery expiration was unlikely), stopping by migrants was infrequent, detection probabilities were consistently high, and unaccounted emigration from the system was highly unlikely. We are substantially less certain about the accuracy of the survival estimates for the lower reaches and the overall survival rate, and we caution against adopting these estimates for White Lake or similar systems. It appears pauses during outmigration were common for our study animals, and stopping was of sufficient frequency and duration that transmitter batteries likely expired prior to exit from the study area. This could have resulted in undetected passage through the river–wetland complex and the drowned rivermouth lake, resulting in underestimated survival.

The simulations that evaluated sample size effects suggested improvements in precision of the survival estimates through the study area could be achieved by increasing the number of tagged lamprey with diminishing returns above a tagging level of 200–300 sea lamprey, at least under the physical conditions of the White River. Even greater levels of tagging may be needed if sea lamprey survival is to be estimated in systems longer in length or where the physical conditions of the stream impede receiver detection probabilities. However, this approach alone likely would not rectify the challenges imposed by the short battery life of microtransmitters coupled with an intermittent and protracted movement schedule. Our simulations support the use of staggered releases across the migration corridor in both space and time to improve CJS model survival estimates for migrating juvenile fish, particularly for downstream sections of the system. This approach came at the expense of slightly worse performance for upstream sections. In practice, the tendency for outmigrating sea lamprey to move intermittently suggests the use of a model that explicitly accounts for temporally varying survival within each of the studied segments (Bunch et al. 2022; Hance et al. 2020). We did not attempt to incorporate time into the CJS model fit to the sea lamprey detection data because of the small sample size and the limited number of spatial locations where receivers were placed. However, given that mortality is inherently a dynamic rate, the use of such models would be reasonable. Survival estimates can also be improved by taking steps to increase detection receiver probability, and we encourage steps to be taken prior to study initiation to ensure detection probabilities are as high as possible through site visits, range testing, and, if necessary, multiple receivers at specific locations. Although not directly simulated, improvements in survival estimates should also accrue by increasing the observation window for any one lamprey by increasing the delay between pings on the transmitter (e.g., a ping interval of 15–20 s would increase tag life to 90+ days). This would necessitate increasing the number of receivers used at any given location and arraying them in strings with overlapping detection radii to ensure a lamprey did not pass through a detection area without issuing a transmission. This approach would also prove useful when experimental subjects are difficult to acquire or enjoy protected status, necessitating the use of fewer subjects.

Regardless of cause, the presence of notable spatial variation in survival across major habitat classes in riverine ecosystems poses a challenge for cost-efficient sea lamprey control in the Great Lakes. The current approach to ranking streams for pesticide treatment (Empiric Stream Treatment Ranking (ESTR)) estimates the abundance of large larvae and assigns a treatment priority based on the anticipated production of newly transformed parasites from those larvae, given stream-specific treatment cost and efficacy (Christie et al. 2003; Jones et al. 2009). The ESTR approach relies, in part, on the assumptions that (a) each larva within a river system has an equal likelihood of survival through metamorphosis and downstream migration and (b) that probability does not vary substantially from river to river throughout the basin. The data reported herein support the hypothesis that survival may vary in relation to the distance traveled from natal burrows to the parasitic feeding grounds and may differ across habitat types. Because the value of killing a larval sea lamprey is proportional to its likelihood of survival in the parasitic phase, the cost-efficiency of the lampricide program would be improved by explicitly accounting for survival within and across infested rivers.