Volume 43, Issue 2 p. 517-538
Open Access

Managing for Atlantic Salmon Smolt Run Timing Variability in a Changing Climate

Danielle M. Frechette

Corresponding Author

Danielle M. Frechette

Maine Department of Marine Resources, Bureau of Sea-Run Fisheries and Habitat, 121 State House Station, Augusta, Maine, 04333 USA

Corresponding author: [email protected]Search for more papers by this author
James P. Hawkes

James P. Hawkes

National Oceanic and Atmospheric Administration, National Marine Fisheries Service, Northeast Fisheries Science Center, 17 Godfrey Drive, Suite 1, Orono, Maine, 04473 USA

Search for more papers by this author
John F. Kocik

John F. Kocik

National Oceanic and Atmospheric Administration, National Marine Fisheries Service, Northeast Fisheries Science Center, 17 Godfrey Drive, Suite 1, Orono, Maine, 04473 USA

Search for more papers by this author
First published: 28 December 2022
Citations: 2


The Gulf of Maine Distinct Population Segment of Atlantic Salmon Salmo salar is listed under the U.S. Endangered Species Act, with dams cited as a key threat to the species. Safe, timely, and efficient passage of emigrating smolts is critical for advancing recovery goals. However, climate change and the earlier (and more variable) onset of smolt migration have challenged the effectiveness of measures designed to protect smolts. Therefore, we used data from four long-term smolt trapping sites in Maine to characterize run duration, identify a standardized smolt run, and develop a predictive model for the initiation of smolt emigration for natural- and hatchery-origin smolts. We combined these data into a basinwide deterministic model that projects the movement of smolts from their point of origin to estuary entry, which we used to (1) evaluate duration and temperature triggers for supplemental spill periods used at hydroelectric dams to protect smolts and (2) investigate how knowledge of smolt emigration dynamics can guide protective measures in rivers where management actions are being developed. Timing of run initiation varied by over 14 d; however, mean air temperature for the first quarter of the year explained over 60% of the variance in the onset of emigration. Initiation temperature was linked with higher-elevation rearing areas where smolts originated and not with downstream dams where temperature was monitored. The shape of the smolt wave varied considerably; thus, our standardized smolt wave captured average dynamics but not the specific shape. Overall, 14-d spill windows are too short. To account for the variable timing and shape of the run, a 54–72-d protection window would be necessary to protect the earliest and latest emigrants in the absence of physical structures (e.g., turbine screens) and preserve the adaptive variation required for populations to respond to climate-driven changes in temperature and hydrology.

Populations of Atlantic Salmon Salmo salar in Maine are listed under the U.S. Endangered Species Act, and the impacts of dams on these fish pose a threat to the species' recovery (National Research Council 2004; Fay et al. 2006; USFWS and NMFS 2018). Although both upstream and downstream passage at dams is essential, losses during the downstream migration of smolts in the springtime are particularly problematic. Estimates of smolt emigration from the Penobscot River—the largest watershed in Maine—from 1970 to 2012 indicated that of 19.7 million smolts stocked, only 7.7 million (39%) entered the ocean, and survival (63–95%) was most variable at dams (Stevens et al. 2019).

Dams can result in direct mortality (e.g., from turbine entrainment or passing over spillways) and can cause delays and injuries that reduce the survival of smolts as they continue their downstream migration to the ocean and beyond (Thorstad et al. 2012; Stich et al. 2015a; Stevens et al. 2019). Such latent effects on survival have been found to be at least comparable to the direct effects of dams in determining smolt migration success (Marschall et al. 2011; Stich et al. 2015a). In the Penobscot River, some of this latent mortality has been linked to a 6% reduction in survival in the estuary for each dam that is passed during out-migration (Stich et al. 2015b). Additionally, delays in run timing caused by dams have been suggested to reduce marine survival by causing smolts to enter the ocean outside of their optimal environmental windows (Russell et al. 2012). There is a rich literature documenting the impacts of dams on both Atlantic Salmon and Pacific salmon Oncorhynchus spp. throughout the world (e.g., Thorstad et al. 2012; Ohms et al. 2022; Waldman and Quinn 2022).

Recovery strategies for the endangered Gulf of Maine (GOM) distinct population segment (DPS) of Atlantic Salmon depend on increasing the number and survival of emigrating smolts that have experienced in-river rearing at some point in their lives. In the GOM DPS, these “natural-origin smolts” (1) originate from natural spawning in river or (2) are individuals that are produced via hatchery spawning and planted in-river at the egg, fry, or parr life stage. Managers have no control over when natural-origin smolts initiate their out-migration (USFWS and NMFS 2018).

The dynamics of smolt emigration have been evaluated at global, regional, and local scales, and these analyses suggest high year-to-year variation in timing (Russell et al. 2012; Otero et al. 2014; Teichert et al. 2020; Vollset et al. 2021). Smolt run timing varies in duration (~4–5 weeks) and time of year depending on location, photoperiod (i.e., day length), and annual environmental triggers (snowmelt, discharge, river temperatures, etc.) across the range of Atlantic Salmon (Baum 1997; Holbrook et al. 2011; Stich et al. 2015a, 2015b; Vollset et al., 2021). There is also a global trend toward earlier run timing because of climate change (Kennedy and Crozier 2010; Otero et al. 2014; Teichert et al. 2020; Gillis et al., in press). Indeed, the initiation of seaward movements is occurring approximately 2.5 d earlier per decade (Otero et al. 2014).

Characterizing the baseline emigration dynamics (timing and duration) for these populations is critical to ensuring that the duration and timing of smolt protection measures at dams are adequate to meet conservation objectives (Thorstad et al. 2012). Additionally, a better understanding of the emigration timing of natural-origin smolts from key production areas can inform smolt stocking programs so that release timing and location of hatchery-stocked smolts can be better synchronized with natural emigration periods and protective measures at dams, thereby minimizing passage delays and mortality at dams and increasing overall survival (Hvidsten and Johnsen 1993; Karppinen et al. 2014). Additional benefits of hatchery- and natural-origin synchrony may include predator swamping and ocean entry timing that is matched appropriately with ocean food sources (Hvidsten et al. 1998; Kennedy and Crozier 2010; Hawkes et al. 2017; Furey et al. 2021).

Smolt out-migration patterns differ between natural- and hatchery-origin smolts (Urke et al. 2013; Stich et al. 2015a, 2015b; Hawkes et al. 2017; Harvey et al. 2020). Hatchery smolts are stocked at discrete locations in the watershed on known dates, whereas natural-origin smolts originate from throughout a watershed as a result of early life stage stocking or natural spawning. Migration speeds of individual smolts may also differ between rearing types (Stich et al. 2015a), as has been observed in the Narraguagus River and other populations (Hansen et al. 1984), although that finding is not universal (McCormick et al. 2014). Because the timing of smolts' interactions with dams is expected to differ by rearing type, efforts to protect smolts from the direct and latent mortality at dams need to account for smolt origin (natural or hatchery).

The effectiveness of measures to protect both natural- and hatchery-origin smolts at dams will depend on the ability to nimbly adapt with shifts in run timing and duration resulting from climate change (Otero et al. 2014; Teichert et al. 2020). Protective measures must start early enough and must be of sufficient duration to ensure that early- and late-migrating smolts pass dams within the protective window to maximize survival at dams, thereby preserving variation in run timing and increasing the potential for adaptation to climate change (Kennedy and Crozier 2010; Otero et al. 2014; Bjerck et al. 2021; Thorstad et al. 2021). These protections also need to include adaptive measures to combat the environmental stochasticity associated with climate change (Thorstad et al. 2012, 2021; Harvey et al. 2020). Recent regional analysis conducted in Norway provided a useful template for developing a tool to provide guidance on expected run timing immediately prior to the onset of smolt movements that (1) uses environmental data available before the onset of emigration (Vollset et al. 2021) and (2) is responsive to climate-driven warming (Teichert et al. 2020).

In this paper, we evaluate the timing and dynamics of Atlantic Salmon smolt emigration from long-term trapping on Maine rivers. We then use these data to evaluate the impact of current regulatory measures that are designed to protect emigrating smolts at dams. We specifically address two requests made by managers: (1) an evaluation of the duration and temperature trigger for a supplementary spill period that is implemented at three main-stem dams on the Penobscot River for the purpose of reducing turbine entrainment of smolts and (2) an investigation of how the knowledge of smolt emigration dynamics could be used to guide development of new protective measures for smolts. Given that the stocking of hatchery-origin smolts remains a core component of Atlantic Salmon recovery actions in the Penobscot River (USFWS and NMFS 2018), managers asked that we evaluate the effectiveness of protective measures for both natural- and hatchery-origin smolts. Our specific objectives were to (1) understand interannual variability in the timing of run initiation; (2) synthesize the available migration data from Maine watersheds to characterize annual emigration patterns and develop a standardized migration pattern; and (3) apply information on run initiation, migration waves, and migration speeds to create informative estimates of smolt emigration timing at a watershed level with daily estimates of occurrence at distinct points of interest (e.g., dams and estuary entry) for both natural- and hatchery-origin smolts in Maine rivers.


Smolt production areas of interest and monitoring

Smolt emigrations in Maine rivers have been studied since the 1960s (Meister 1962; Baum and Jordan 1982). Most data are from coastal rivers or tributaries of the Penobscot and Kennebec rivers. Based on Penobscot River smolt trapping studies conducted from 2000 to 2005, smolts migrated in the Penobscot River between late April and early June, with a peak in early May (reported by Fay et al. 2006). Under the critical habitat designation for the GOM DPS of Atlantic Salmon, rearing habitat was quantified through a combination of modeled and measured habitat (NMFS 2009). As not all habitat has been directly measured, the number of modeled habitat units (where one habitat unit is equivalent to 100 m2) provides a useful metric with which to estimate potential smolt production from a given tributary.

Within the Penobscot River watershed, we focused on three major tributaries (hereafter, “production areas”) of management concern for predicting smolt emigration timing: (1) Piscataquis River, (2) East Branch Penobscot River, and (3) Mattawamkeag River. The Piscataquis River has a watershed area of 377,853 ha and 36,748 modeled rearing habitat units. Smolts migrating from the Piscataquis River production area must pass three existing dams (Guilford, Moosehead, and Brown's Mill dams) and the decommissioned Howland Dam before reaching the main-stem Penobscot River (Figure 1). The East Branch Penobscot River has a watershed area of 289,561 ha, has 22,198 modeled rearing habitat units, and is upstream of the Mattaceunk Dam (Figure 1). The Mattawamkeag River has a watershed area of 390,631 ha and 32,496 modeled rearing habitat units and joins the Penobscot River upstream of West Enfield Dam. With an estimated smolt production potential of 1–3 smolts/rearing habitat unit (Baum 1997; Fay et al. 2006), the total production of smolts from these three production areas combined, if fully occupied and productive, would range from 91,422 to 274,326 smolts. Once they enter the main-stem Penobscot River, smolts originating from the three production areas will either (1) continue down the main-stem Penobscot River and encounter one additional dam (Milford Dam) or (2) emigrate through a side channel that is parallel to the main stem, where they encounter two additional dams before entering the free-flowing lower Penobscot River en route to the estuary (Figure 1).

Details are in the caption following the image
Map of temperature monitoring stations in Maine used for regression development (interpolated from North American Regional Reanalysis data at Whitefield, Dover-Foxcroft, Medway, and Cherryfield; open circles); temperature monitoring stations used for predicting the day of year of 25% capture for Atlantic Salmon smolts from the model (measured air temperature from Millinocket Municipal Airport, Houlton International Airport, Dover-Foxcroft Wastewater Treatment Plant, and Farmington Wastewater Treatment Plant [closed circles] and interpolated data for East Machias [open circle]). Smolt trapping sites used for model development and model testing are indicated (gray stars). Dam locations referenced in this study are denoted by numbered shields: (1) Milford, (2) West Enfield, (3) Mattaceunk, (4) Howland, (5) Brown's Mill, (6) Moosehead, and (7) Guilford dams in the Penobscot River drainage and (8) Lockwood, (9) Hydro-Kennebec, (10) Shawmut, and (11) Weston dams in the Kennebec River drainage. Solid triangles indicate other dams in each watershed that may impede salmon movements.

Currently, the only actively managed production area in the Kennebec River watershed is the Sandy River, which contains 43,137 modeled rearing habitat units. Smolts originating in the Sandy River must pass four main-stem dams on the Kennebec River to reach the estuary: Lockwood, Hydro-Kennebec, Shawmut, and Weston dams (Figure 1). Smolt production potential in the Sandy River is between 43,137 and 129,411 if the habitat is fully occupied.

The four salmon production areas described above are important to managers because of their potential to produce large numbers of smolts. However, direct monitoring of smolts in these areas has been very limited. In the Penobscot River watershed, recent monitoring has occurred only in the Piscataquis River (2008–2015); historical data from a trap at Mattaceunk Dam were available but more dated (1988–1990, 1993–1995, and 1997). The Sandy River production area had its first comprehensive (full-season) monitoring in 2021. Longer and more recent time series of smolt data were available from the Narraguagus River (1997–2019) and Sheepscot River (2001–2019). More limited smolt migration data were available from sites on the East Machias River and Pleasant River, a tributary of the Piscataquis River.

It would be ideal to directly monitor smolt emigration dynamics in all four production areas annually to understand run timing and mitigate risk to emigrating smolts. However, with limited resources, direct smolt monitoring is not always feasible. Even when resources are available, the spring high-flow conditions that often occur during the smolt migration period can prevent trapping activities or result in gaps in data collection. Because regional models have performed well for predicting smolt emigration dynamics (e.g., Vollset et al. 2021), we used fish captures at four long-term smolt trapping stations in Maine to (1) characterize migration patterns and (2) build a predictive run timing model.

Characterizing smolt migration timing and patterns in Maine

Fish collection platforms at a dam bypass device and rotary screw traps (RSTs; E. G. Solutions) were used to collect smolt run timing data. Operation of these collection platforms was consistent from year to year in that deployment was initiated shortly after ice-out to capture the full extent of the smolt run. On the Penobscot River, out-migrating Atlantic Salmon smolts were collected at a downstream fishway bypass located at the Mattaceunk Hydroelectric Project (Federal Energy Regulatory Commission [FERC] Project 2520). Smolts bypassed the dam and were held within the holding/sorting facility, where they were collected daily for enumeration and measurement. Smolt trapping on the Piscataquis, Narraguagus, and Sheepscot rivers was conducted using 1.5-m RSTs. These passive fish collection devices have minimal effects on Atlantic Salmon smolts, as the collection cone directs fish and disperses the energy of the river flow while funneling out-migrating smolts into a rear holding box (Music et al. 2010). Once captured, smolts remained in holding boxes overnight prior to daily collection and processing by biologists, at which time they were counted, measured, and sampled before release.

For each smolt trapping site, we used the daily catch data to calculate the run duration and four metrics describing run timing: the day of year (DOY) that 5, 25, 50, and 95% of the smolt run was captured at each site. Only years for which smolt trapping data sets were considered complete were included in the analysis (Figure 2). Annual data sets were considered complete if (1) the start and end of the run were captured, as indicated by 3–5 d of zero capture before the first fish was captured and after the last fish was captured; and (2) no large gaps in trapping effort occurred because of high river discharge or other factors.

Details are in the caption following the image
Atlantic Salmon smolt run timing on the Penobscot (Mattaceunk Dam), Narraguagus, Piscataquis, and Sheepscot rivers, Maine. The box dimensions indicate the 25th and 75th percentiles, the whiskers represent the 5th and 95th percentiles, and the horizontal lines within the boxes denote the medians. Blue boxes indicate data that were used to parameterize the regression model; gray boxes represent data that were omitted from analysis because the data were incomplete. Gaps in data are years in which no sampling occurred.

Predictive run timing model

We used a regression approach to predict DOY for the four run timing metrics (5, 25, 50, and 95% smolt capture) from predictor variables that would be available annually on April 1. We chose this date to provide guidance for managers to annually predict the onset of smolt migration in near-real time. Air temperature (°C), water temperature (in accumulated thermal units [ATUs]), and discharge have all been identified as variables influencing the timing and dynamics of smolt emigration (e.g., Zydlewski et al. 2005; Vollset et al. 2021), and they were included as predictor variables in our modeling approach, along with trap elevation and cumulative drainage area (CDA).

Direct measurements of historic air temperature data were not available for all combinations of years and locations; therefore, we obtained air temperature data from the National Centers for Environmental Prediction (NCEP) North American Regional Reanalysis—a high-resolution, long-term climate data set for North America (Mesinger et al. 2006). The NCEP reanalysis data were from the Physical Sciences Laboratory (National Oceanic and Atmospheric Administration [NOAA], Earth System Research Laboratories; obtained on July 30, 2021, from https://psl.noaa.gov/data/gridded/data.narr.html). The NCEP data were available as daily mean temperature, from which we computed two summary metrics: the mean air temperature (Tmean) and cumulative air temperature (Tcum) for the first quarter of the year (sensu Vollset et al. 2021), which extended from January 1 to March 31 (in standard years) or March 30 (in leap years). We defined Tmean as the mean air temperature from day 1 to day 90. We calculated Tcum for the first 90 d of the year by using a threshold temperature of 0°C, where Tcum was the sum of all mean daily temperatures greater than 0°C (i.e., T cum = T > 0 for day 1 to day 90). We processed air temperature data to obtain annual Tmean and Tcum using the tidyverse package (Wickham et al. 2019) with R version 4.1.0 (R Core Team 2021).

Water temperature was described in ATUs because of previously published links between this metric and smolt emigration timing (Zydlewski et al. 2005; Stich et al. 2015a). We used a threshold of 0°C to calculate ATUs from day 1 (January 1) to day 90 for the years and trap sites with available water temperature data. Trap elevation (m) was obtained from Google Earth Pro to account for any potential differences in winter air temperature regime between traps located nearer to sea level (Sheepscot River: 4 m; Narraguagus River: 22 m) and those located at higher elevations (Piscataquis River: 126 m; East Branch Penobscot River: 72 m). Cumulative drainage area in square kilometers was included to account for differences in upstream drainage area that could be contributing to smolt production and was obtained from the U.S. Geological Survey (USGS) gauging station situated nearest to each trap site (Sheepscot River: 01038000; Piscataquis River: 01031300; East Branch Penobscot River: 01029500; Narraguagus River: 01022500).

Winter–spring volumes (WSVs; normalized by drainage area, with units of cm) and winter–spring center-volume dates (WSCVDs; DOY) were computed following the methods of Dudley et al. (2017). The WSV was calculated as the sum of runoff volume based on daily streamflow data collected at the USGS gauge located in proximity to each smolt trapping station, and the WSCVD was calculated as the date that half of the streamflow volume passed each gauging station. These streamflow metrics were computed for three different time intervals (beginning on January 1 and ending on March 1, March 15, or March 31) for each smolt trapping location and year.

Exploratory plots of the data indicated a linear relationship between the candidate run timing metrics and each predictor variable, with the exception of elevation. We regressed each of the four run timing metrics (i.e., DOY of 5, 25, 50, or 95% smolt capture) against the predictors Tmean, Tcum, ATUs, elevation, CDA, WSV, and WSCVD in a set of a priori defined combinations using forward stepwise selection (Crawley 2007; James et al. 2021). In each step, we selected the linear model that maximized the adjusted R2 (James et al. 2021). Candidate models contained only predictor variables that were correlated with an R-value less than 0.60 to avoid problems with multicollinearity (Dormann et al. 2013). Residuals were plotted and visually examined to test the assumptions of normality and homogeneity of variance and to identify any influential data points. Model fit was assessed using leave-one-out cross validation (Hastie et al. 2009). Regression analyses were conducted using the lm function, and diagnostic plots were generated using the plot function in R version 4.1.0 (R Core Team 2021). The run timing metric that was associated with the greatest R2 value was retained as best describing the onset of migration.

We used smolt trapping data from RST sites on the Pleasant and East Machias rivers (Figure 1) to assess the efficacy of the selected regression model in predicting the onset of emigration. Only years when smolt captures exceeded 50 individuals were included in this analysis. We predicted the onset of emigration in the Pleasant River by using measured air temperature from the Dover-Foxcroft Wastewater Treatment Plant (station USC00171975; https://mco.umaine.edu/data_daily/; accessed February 2, 2022). Measured air temperature was not available for the East Machias River; therefore, we used interpolated data from the NCEP North American Regional Reanalysis (accessed January 27, 2022; Mesinger et al. 2006) to predict emigration onset.

We also investigated the variability in the estimated onset of emigration to help inform whether and how much of a buffer should be applied to smolt protection windows to ensure that the tails of the run are adequately protected. To do this, we applied the selected regression model to measured air temperature (i.e., Tmean) obtained for the years 2000–2021 for each of the four production areas of interest (Piscataquis, East Branch Penobscot, and Mattawamkeag rivers in the Penobscot River drainage; Sandy River in the Kennebec River drainage) and computed the spread in the model predictions.

Systemwide smolt movement model

Smolt out-migration can be described as a “wave” moving downstream, beginning with an ascending limb of early migrants, followed by an increase in the daily number of migrants to a peak, and concluding with a descending limb of late migrants. We built a deterministic, predictive model to project the movement of the wave downstream through dams and other ecological points of interest (e.g., estuary entry), hereafter referred to as the systemwide “smolt movement model” (Figure 3). The inputs for the smolt movement model are (1) the predicted onset of emigration generated from the run timing model described in the previous section, (2) a standardized smolt wave for either natural- or hatchery-origin smolts, (3) estimated habitat-specific migration speed (from the literature), and (4) river-specific migration distance from the start of downstream movement to estuary entry.

Details are in the caption following the image
Flow diagram detailing the inputs and application of the systemwide migration model for two Atlantic Salmon smolt production areas and one hatchery smolt stocking location (DOY = day of year; Tmean = mean air temperature). Key model inputs are the run timing model (Table 2); standardized smolt waves (Figure 4); spatially explicit distances through free-flowing reaches, head ponds, and dam reaches; and habitat-specific migration speed (km/h) from Stich et al. (2015a; F = free-flowing reach migration speed, H = head pond migration speed, and D = dam reach speed).

We developed the standardized smolt waves by using catch data from the Narraguagus River RST, which had the longest and most complete time series of available data for natural-origin (1997–2019) and hatchery-origin (2008–2012) smolts. Given the size of the watershed monitored, consistent wild production, and routine fry stocking, this monitoring site is an appropriate proxy for nearby watersheds of similar size. Hatchery smolts had the adipose fin removed before stocking, so they were visually distinguishable from natural-origin smolts at the RST. To obtain the natural-origin standardized wave, we calculated the daily catch as a percentage of the total seasonal catch for each year (Figure 4). The resulting annual distributions were then centered across years by aligning the median catch day. Next, the average daily run proportion was averaged across the entire time series to generate the natural-origin standardized smolt wave, which represents a population leaving a discrete production area (Figure 3). Because hatchery-stocked smolts were introduced into the Narraguagus River on known dates and locations, capture dates at the RST were back-calculated to stocking dates to develop the standardized hatchery wave (Figure 4). Because the stocking location was typically about 20 km upstream of the smolt trapping location, capture days were adjusted annually to reflect initiation of the wave at the stocking site by using the migration speeds of hatchery fish in free-flowing river segments (Stich et al. 2015a).

Details are in the caption following the image
Atlantic Salmon smolt waves in the Narraguagus River from 1997 to 2019. The top panel shows the percent of natural-origin smolts captured by day of year: the solid orange line represents an early run start (April 16, 2010), and the dashed orange line represents the latest start (May 7, 2019); the solid blue line represents the shortest run duration (2015: 20 d), and the dashed blue line represents the longest run duration (2000: 52 d); and all other years are shown in light gray. The middle panel presents the proportion of natural-origin smolts captured per day for the duration of the run. Framing years (2000, 2010, 2015, and 2019) use the same scheme (colors and line types) as the top panel, and the bold black line represents the calculated average smolt wave. The bottom panel represents the average catch rates of hatchery-stocked smolts, with day 1 as the stocking date for Narraguagus River monitoring (2008–2012).

Once the standardized waves were developed, they were aligned with the predicted onset of emigration from the run timing model or the hatchery stocking date. Each daily cohort from a production area or stocking event was then “moved downstream” using the measured reach length and origin-specific movement rates by reach type. The distance traveled by each daily cohort to a point of interest (dam locations or estuary entry) was determined by spatially explicit migration routes and distances through free-flowing reaches, impoundments (head ponds), and dam reaches (Supplementary Material available in the online version of this article). Migration speed (km/h) from published estimates generated for the Penobscot River (Stich et al. 2015a) was assigned to each daily cohort based on origin (hatchery versus natural) and location within the watershed (free-flowing reach, head pond, or dam reach). For example, we assigned the free-flowing migration speed (from Stich et al. 2015a) to smolts when they occupied freely flowing river reaches. Once smolts entered a dam head pond, we set the migration speed to the head pond speed until smolts were within 1 km of the dam, at which point we assigned them the dam speed (Stich et al. 2015a).

We designed data frames for the Penobscot and Kennebec River watersheds that encompassed all major subdrainages as distinct production areas (Figure 1) and ended near the head of tide. We built the smolt movement model for production areas in the Penobscot and Kennebec rivers and made it available in a spreadsheet format that managers can use in real time. The final product of the smolt movement model is a tool that can be adjusted/calibrated at day 90 of each year to estimate the migration timing of standardized smolt waves and can provide guidance to managers on likely fish passage times at discrete areas of interest, such as dams or estuary entry (Figure 3). In the following sections, we use three case studies to demonstrate the utility of the systemwide smolt movement model and address requests by managers.

Case study 1: evaluating current protective measures across penobscot river production areas

We evaluated the timing of protective measures relative to the year-specific standardized smolt waves for Penobscot River dams in 2019–2021. Three distinct time windows have been used to protect Atlantic Salmon smolts during downstream migration through FERC-licensed dams in the Penobscot River watershed. These projects are required to have dedicated downstream fish passage systems that operate from April 1 to June 30. In addition to operating downstream passage facilities, operators must spill water over the dams or through gate structures during the spring to reduce turbine entrainment, with the goal of providing increased protection for the peak of the smolt run. The requirement to spill additional water to protect Atlantic Salmon is contained in the FERC licenses to operate the projects (NMFS 2012). The dam operator was required to initiate spillage for 14 d at Milford, Orono, Stillwater, and West Enfield dams once the ambient river temperatures (measured at Milford Dam) reached 10°C. The protective measure at Mattaceunk Dam was an increased flow to a log sluice for 21 d once a trigger temperature of 10°C was observed at the facility. The FERC license for the Brown's Mill Dam on the Piscataquis River requires the operation of a downstream fishway smolt bypass from ice-out (~April 1) until June 30; because of the high survival at this site, no supplemental spill is required (e.g., Molina-Moctezuma et al. 2021).

To assess these protective windows (i.e., operation of downstream fish passage systems and 14-d supplemental spill periods), we calculated the proportion of the standard smolt waves from the Piscataquis, East Branch Penobscot, and Mattawamkeag River production areas that migrated past Mattaceunk, West Enfield, Brown's Mill, and Milford dams when the measures were in place. We considered only the passage route through the main stem and Milford Dam because most smolts use this route (Holbrook et al. 2011; Molina-Moctezuma et al. 2021). For illustrative purposes, we used natural production of 2 smolts/rearing habitat unit as migrating populations (Baum 1997; Fay et al. 2006). While site-specific information was available for all 3 years and sites, we illustrate the outcomes at four primary facilities in 2021 (Mattaceunk, West Enfield, Brown's Mill, and Milford dams) and provide a summary of protective coverage provided by the supplemental spill at Milford Dam.

Case study 2: modeled protective measures compared with observed catches in the Sandy River

We used the Sandy River production area as a case study to compare the output of the systemwide smolt movement model with active smolt monitoring. We first projected the 2021 smolt wave by using a systemwide migration model for the Kennebec River, where the inputs were the run timing model applied to the Sandy River production area and the standardized natural-origin smolt wave. We set the first day that smolts arrived at a dam as the beginning of a protective window.

We then applied the systemwide migration model to the 2021 smolt trapping data for the Sandy River. The first day on which smolts were caught in the Sandy River RST was set as the starting day for the natural-origin smolt wave. We projected the movement of daily cohorts (prorated by the population estimate for 2021) through the four main-stem dams in the Kennebec River to estimate the percentage of the smolt run passing within and outside of the protective window.

Case study 3: Piscataquis River natural/hatchery smolt waves and all stocking events in 2019

In recent years, stocking of hatchery smolts in the Penobscot River has primarily occurred downstream of Milford Dam to maximize the in-river survival of smolts to the estuary; between 2019 and 2021, smolts were stocked above existing dams only during 2019. We applied the smolt migration model to the 2019 stocking data to compare (1) the arrival timing of both natural- and hatchery-origin smolt waves at two dams (Brown's Mill and Milford) and (2) the timing of estuary entry for all smolt waves. In 2019, a total of 596,000 smolts was stocked on 10 dates between April 22 and May 2 at three locations within the watershed: upstream of and downstream of Milford Dam (labeled as Dam 1 in Figure 1) and upstream of Moosehead Dam (labeled as Dam 6 in Figure 1). Stocking locations, dates, and numbers of smolts were obtained from the U.S. Fish and Wildlife Service (Oliver Cox, Green Lake National Fish Hatchery, personal communication).


Characterizing Smolt Migration Timing and Patterns in Maine

Monitoring of smolt migrations on the Narraguagus, Penobscot (Mattaceunk Dam), Piscataquis, and Sheepscot rivers was consistent across most years (Figure 2). Capture dates exhibited variability among years and sites, but the median date of capture generally occurred within a 10-d window (Figure 2). When using the first and last captures to calculate run duration, the minimum duration was at least 3 weeks, while the maximum duration was between 42 and 58 d (Table 1). When run duration was computed as the central 90% of the smolt run, the minimum duration was approximately 14 d and the maximum duration was between 22 and 35 d (Table 1).

TABLE 1. Data from four long-term Atlantic Salmon smolt trapping sites in Maine, summarized as the timing and duration of smolt captures. Earliest and latest day of first and last captures (day of year) are presented with the median, mean, minimum (min), and maximum (max) run duration, where run duration is the number of days elapsed between the first and last smolt captures. Mean day of year (with SE in days) that 5% and 95% smolt capture occurred is presented with the median, mean, minimum, and maximum run duration, where run duration is the number of days elapsed between the dates of 5% and 95% smolt capture (i.e., the amount of time required for 90% of the smolt run to pass the trapping site).
Variable Statistic Trap site
Mattaceunk Dam Narraguagus River Piscataquis River Sheepscot River
Number of trapping years evaluated 7 20 7 17
First day of capture Earliest/latest 110/139 103/128 105/127 101/122
Last day of capture Earliest/latest 168/182 144/164 141/157 142/156
Run duration (first capture to last capture) Median 56 34 29 34
Mean 52 35 30 33
Min (year) 36 (1997) 20 (2015) 22 (2015) 21 (2007)
Max (year) 58 (1988) 52 (2000) 42 (2008) 44 (2010)
Day of 5% smolt capture Mean (SE) 137 (3) 123 (1) 125 (3) 119 (1)
Day of 95% smolt capture Mean (SE) 159 (2) 141 (1) 140 (2) 140 (1)
Run duration (5–95% smolt capture) Median 21 18 15 20
Mean 22 19 15 21
Min (year) 14 (1997) 14 (2001) 9 (2011) 15 (2007)
Max (year) 34 (1990) 25 (2010) 22 (2010) 27 (2011)

Predictive Run Timing Model

The Tmean was highly correlated with ATUs, elevation, and CDA, whereas Tcum was highly correlated with ATUs (R > 0.60). Elevation was both non-linear and highly correlated with the other variables and was not considered further. Models including CDA, ATUs, WSV, or WSCVD explained very little of the variation in run timing (Table 2; Supplementary Table 1). The relationship between Tmean and DOY was significant and linear for each of the four run timing indicators evaluated (i.e., DOY of 5, 25, 50, and 95% smolt capture), and in each case it was the model that maximized adjusted R2 (Crawley 2007; James et al. 2021). Plots of model residuals indicated that the assumptions of normality and homogeneity of variance were adequately met. Data points for Mattaceunk Dam (in 1997) and the Piscataquis River (in 2015) were deemed influential (Cook's distance ~0.5); however, model fit was considered adequate based on leave-one-out cross validation (R2 = 0.6556). The Tmean during the first 90 d of the year was identified as the minimal adequate model for predicting the DOY that 5, 25, and 50% of the smolt run was captured (Table 2; Supplementary Table 1). In each case, Tmean predicted over 60% of the variance in the run timing metric; however, Tmean predicted less than 50% of the variation in the DOY of 95% smolt capture. The linear regression model predicting the DOY that 25% of smolts were captured (equation 1) explained 66% of the variance in capture date:
DOY = 123.72 1.96 T mean . (1)
This model was used to predict the onset of smolt migration and to anchor the standardized natural-origin smolt wave in the systemwide smolt movement model (Figure 3).
TABLE 2. Model selection process for predicting the day of year (DOY) of 25% capture for Atlantic Salmon smolts by using linear regression (Tmean = mean air temperature; Tcum = cumulative air temperature; ATU = accumulated thermal units; wsv = winter–spring volume; wscvd = winter–spring center-volume date). Each a priori defined model is presented with estimated model coefficients and associated SEs, t-statistic with P-value, residual square error (RSE) and associated df, multiple R2, adjusted R2 (used for model selection), and F-ratio with associated df and P-value.
Run timing metric Model Coefficient Estimate SE t P RSE (df) Multiple R2 Adjusted R2 F (df) P
25% smolt capture DOY ~ Tmean Intercept 123.72 0.85 145.25 <2 × 10−16 4.467 (46) 0.6556 0.6481 87.58 (1, 46) 0.0000
T mean −1.96 0.21 −9.36 0.00000
DOY ~ Tcum Intercept 135.91 1.37 99.41 <2 × 10−16 5.709 (46) 0.5061 0.4953 47.13 (1, 46) 0.0000
T cum −0.10 0.01 −6.87 0.00000
DOY ~ ATU Intercept 129.35 0.98 131.42 <2 × 10−16 3.821 (28) 0.4126 0.3916 19.67 (1, 28) 0.0001
ATU −0.09 0.02 −4.44 0.00013
DOY ~ wsv_cm_Mar1 Intercept 128.37 1.62 79.15 <2 × 10−16 4.999 (39) 0.0706 0.0468 2.96 (1, 39) 0.0931
wsv_cm_Mar1 −0.21 0.12 −1.72 0.09310
DOY ~ wsv_cm_Mar15 Intercept 129.67 1.61 80.62 <2 × 10−16 4.780 (39) 0.1502 0.1228 6.90 (1, 39) 0.0123
wsv_cm_Mar15 −0.24 0.09 −2.63 0.01230
DOY ~ wsv_cm_Mar31 Intercept 130.66 1.76 74.34 <2 × 10−16 4.686 (39) 0.1834 0.1624 8.76 (1, 39) 0.0052
wsv_cm_Mar31 −0.23 0.08 −2.96 0.00522
DOY ~ wscvd_jd_Mar1 Intercept 131.36 2.30 57.11 <2 × 10−16 4.814 (39) 0.1382 0.1161 6.26 (1, 39) 0.0167
wscvd_jd_Mar1 −0.20 0.08 −2.50 0.01670
DOY ~ wscvd_jd_Mar15 Intercept 131.39 1.97 66.82 <2 × 10−16 4.677 (39) 0.1867 0.1659 8.96 (1, 39) 0.0048
wscvd_jd_Mar15 −0.15 0.05 −2.99 0.00478
DOY ~ wscvd_jd_Mar31 Intercept 130.95 2.39 54.81 <2 × 10−16 4.887 (39) 0.1120 0.0893 4.92 (1, 39) 0.0324
wscvd_jd_Mar31 −0.10 0.05 −2.22 0.03240
DOY ~ Tmean + wsv_cm_Mar1 Intercept 124.07 1.77 70.00 <2 × 10−16 4.278 (38) 0.3370 0.3021 9.66 (2, 38) 0.0004
T mean −1.25 0.32 −3.91 0.00037
wsv_cm_Mar1 0.01 0.12 0.08 0.93967
DOY ~ Tmean + wscvd_jd_Mar1 Intercept 126.77 2.38 53.31 <2 × 10−16 4.206 (38) 0.3590 0.3253 10.64 (2, 38) 0.0002
T mean −1.10 0.30 −3.62 0.00086
wscvd_jd_Mar1 −0.09 0.08 −1.15 0.25927
DOY ~ Tmean + wscvd_jd_Mar15 Intercept 127.24 2.13 59.62 <2 × 10−16 4.135 (38) 0.3752 0.3423 11.41 (2, 38) 0.0001
T mean −1.03 0.31 −3.39 0.00166
wscvd_jd_Mar15 −0.07 0.05 −1.53 0.13521
DOY ~ Tmean + wscvd_jd_Mar31 Intercept 126.57 2.36 53.59 <2 × 10−16 4.215 (38) 0.3562 0.3223 10.51 (2, 38) 0.0002
T mean −1.13 0.30 −3.80 0.00052
wscvd_jd_Mar31 −0.04 0.04 −1.07 0.29278

Run Timing Model Evaluation

Four years of natural-origin smolt emigration data were available for the Pleasant River (2004–2008), and 7 years of data were available for the East Machias River (2013–2019). In general, the predicted DOY of 25% smolt emigration (equation 1) was within 5 d of the actual date of 25% emigration for both the Pleasant River and East Machias River RST sites. The greatest difference between observed and predicted days of 25% emigration was 9 d for the Pleasant River and 7 d for the East Machias River (Table 3).

TABLE 3. Predicted and observed day of year (DOY) of 25% capture for natural-origin Atlantic Salmon smolts trapped on the Pleasant River (2004–2008) and East Machias River (2013–2019) by the Maine Department of Marine Resources and National Oceanic and Atmospheric Administration Fisheries, presented with the difference between observed and predicted DOYs. Predicted values were obtained using equation (1). Only years with smolt captures greater than 50 individuals were considered in this analysis.
River Year Number of fish Observed DOY Predicted DOY Difference (predicted - observed)
Pleasant 2004 58 126 135 9
2006 117 123 128 5
2007 102 137 135 −2
2008 60 135 131 −4
East Machias 2013 141 131 128 −3
2014 159 133 132 −1
2015 74 130 137 7
2016 206 130 127 −3
2017 260 124 129 5
2018 198 126 128 2
2019 220 132 132 0

For the period 2000–2021, we observed considerable variability in the estimated day of 25% emigration (Figure 5) in all four production areas of interest (Piscataquis, East Branch Penobscot, and Mattawamkeag rivers in the Penobscot River drainage; Sandy River in the Kennebec River drainage). For all sites combined, the day of 25% emigration ranged from DOY 127 to 146 (median DOY = 136). When considering each production area independently, the median predicted date of 25% emigration was DOY 135 (Sandy River), DOY 136 (East Branch Penobscot River), DOY 137 (Piscataquis River), and DOY 139 (Mattawamkeag River). The maximum difference in the predicted day of 25% emigration across the three Penobscot River production areas was 5 d, which occurred during 2001 (DOY 137 in the Piscataquis River, DOY 142 in the Mattawamkeag River, and DOY 138 in the East Branch Penobscot River) and during 2008 (DOY 136 in the East Branch Penobscot River, DOY 141 in the Mattawamkeag River, and DOY 138 in the Piscataquis River).

Details are in the caption following the image
Variability in the predicted day of year (DOY) of 25% emigration for the four Atlantic Salmon smolt production areas of interest from air temperature measured at the Millinocket Municipal Airport (East Branch Penobscot River), Houlton International Airport (Mattawamkeag River), Dover-Foxcroft Wastewater Treatment Plant (Piscataquis River), and Farmington Wastewater Treatment Plant (Sandy River). The box dimensions indicate the 25th and 75th percentiles, the whiskers represent the 5th and 95th percentiles, and the horizontal lines within the boxes denote the medians.

Systemwide Smolt Movement Model

The natural-origin smolt run in the Narraguagus River exhibited a unimodal distribution in some years, whereas in other years, a bimodal distribution occurred. The standardized natural-origin smolt wave developed using these catch data was unimodal and had a duration of 54 d, where 25% of the total catch occurred on day 21 (Figure 4). The wave progressed slowly for the first 10 d before movements increased; from day 20 to day 27, daily catches were at least 5% of the total run, representing the core of the run (Figure 4). The tail of the run had a more gradual slope and was longer (28 d) than the start of the wave.

Catches of hatchery-stocked smolts in the Narraguagus River indicated a very different shape and duration of emigration compared with natural-origin smolts (Figure 4). The first captures of hatchery smolts in the RST typically occurred on the day after stocking. In most years, smolts were stocked approximately 20 km upstream of the trapping site; capture data indicate that these fish moved downstream on the day of stocking, given an average movement rate of 1.09 km/h (Stich et al. 2015a). The duration of the hatchery smolt run was 19–23 d, resulting in a 3-d running average of 23 d (Figure 4). The hatchery smolt wave progressed more rapidly than the natural-origin smolt wave, with daily catches of over 10% of the run in the first 4 d postrelease and over 5% of the run for the next 5 d.

Case Study 1: Current Protective Measures Across Penobscot River Production Areas

Run initiation varied among years (2019–2021) and among production areas due to differences in air temperature. The DOY of 25% emigration occurred earlier in the East Branch Penobscot River than in the Mattawamkeag or Piscataquis River in all 3 years (Table 4) but only by 1–3 d. The proportion of the run that passed Milford Dam during the 14-d supplemental spill window at the dam was 0% in 2021 and 68% in 2020 (Table 4).

TABLE 4. Mean air temperature (Tmean) for the first quarter measured for three Atlantic Salmon smolt production areas in the Penobscot River watershed, with the predicted day of year (DOY) of 25% smolt out-migration (generated using the equation DOY = 123.72–1.96 × Tmean), the predicted run timing at Milford Dam (generated using the smolt wave simulation with 10,000 natural-origin smolts originating in each production area), the actual protection window at Milford Dam, and predicted percentage of the smolt run that was protected for the years 2019–2021 (MA = Municipal Airport; IA = International Airport; WWTP = Wastewater Treatment Plant).
Year Smolt production area Air temperature monitoring location Tmean (°C; days 1–90) Predicted DOY of 25% smolt emigration Predicted arrival at Milford Dam (DOY) Milford Dam protection window start (DOY) Percentage of run passing during protection window
2019 East Branch Penobscot River Millinocket MA −7.56 139 120–173 127 25
Mattawamkeag River Houlton IA −9.12 142 123–177
Piscataquis River Dover-Foxcroft WWTP −8.65 141 123–177
2020 East Branch Penobscot River Millinocket MA −4.48 133 114–163 136 68
Mattawamkeag River Houlton IA −6.43 136 117–171
Piscataquis River Dover-Foxcroft WWTP −5.19 134 116–165
2021 East Branch Penobscot River Millinocket MA −4.51 133 114–163 102 0
Mattawamkeag River Houlton IA −5.49 134 115–169
Piscataquis River Dover-Foxcroft WWTP −5.05 134 116–165

During 2021, the operation of the downstream bypass at Brown's Mill Dam on the Piscataquis River was in effect for the entire Piscataquis River standard wave (Figure 6). At Mattaceunk Dam, the increased flow to the log sluice started on April 29, 2021 (DOY 119), and was in effect for about 76% of the East Branch Penobscot River standard wave (Figure 6). All fish passing outside this window did so after the 21-d period (representing the later quarter of the smolt migration).

Details are in the caption following the image
Simulated 2021 migration waves for the East Branch Penobscot River (yellow), Piscataquis River (green), and Mattawamkeag River (orange) Atlantic Salmon smolt production units as the smolts transited four dams on their seaward migration. Shaded boxes with arrows indicate the period during which supplemental spill measures were in place at Mattaceunk, West Enfield, and Milford dams. No supplemental spill was required at Brown's Mill Dam.

Initiated by the 10°C trigger temperature (measured at Milford Dam), the 14-d supplemental spill window at West Enfield and Milford dams started on April 12, 2021 (DOY 102), and ended on April 25, 2021 (DOY 115). At West Enfield Dam, the supplemental spill window ended 2 d after the arrival of the East Branch Penobscot River wave (on DOY 113) and 1 d before the arrival of the Mattawamkeag River wave (on DOY 116). Consequently, only 0.05% of the East Branch Penobscot River standard wave and none of the Mattawamkeag River standard wave passed West Enfield Dam during the 14-d supplemental spill window. The first fish from each production area arrived at Milford Dam after the 14-d supplemental spill window had ended (Figure 6).

Case Study 2: Modeled Protective Measures Compared with Observed Catches in the Sandy River

In 2021, Tmean for the Sandy River (measured at the Farmington Wastewater Treatment Plant; Figure 1) was −4.01°C and the predicted day of 25% smolt capture was May 11 (DOY 131). Aligning the predicted DOY of 25% emigration (DOY 131) with day 21 of the 54-d natural-origin standard wave allowed us to estimate that the Sandy River smolt wave started on April 20, 2021 (DOY 110), and ended on June 12, 2021 (DOY 163).

Inputting these values into the systemwide smolt movement model for the Kennebec River (with the protective window duration set at 54 d for each dam) produced protective window recommendations that progressed moving downstream, starting with Weston Dam from DOY 111 to DOY 164 (Figure 7). Due to their close proximity, the suggested protective windows at Shawmut, Hydro-Kennebec, and Lockwood dams would be DOY 112–165.

Details are in the caption following the image
Sandy River 2021 simulated protection window, with the top panel (black bars) depicting the simulated Atlantic Salmon smolt wave and timing upon arrival at Weston Dam on the Kennebec River (DOY = day of year). The bottom panel (stippled bars) indicates the smolt trap-based extrapolation of smolt arrivals at Weston Dam in 2021. Arrows indicate the placement of a protective window generated using the simulation start (DOY 111) and end (DOY 164) days.

These protective window recommendations were then applied to daily catch ratios at the Sandy River smolt trap, which were expanded to represent the 2021 smolt population estimate of about 13,500 smolts. The date of 25% smolt capture in the Sandy River RST was May 2 (DOY 122), 9 d earlier than the model prediction of DOY 131 (Figure 7). Even with a relatively rapid onset of the smolt run and highly variable daily catch rates, approximately 99% of the Sandy River smolt wave would have passed all four of the dams during the suggested 54-d protective window (Figure 7). Expansion of this window by 6 d earlier and 5 d later would have protected the entire run as estimated by trap catch.

Case Study 3: Piscataquis River Natural/Hatchery Smolt Waves and All Stocking Events in 2019

Comparison of the natural-origin smolt wave with the hatchery-origin smolt wave in the Piscataquis River indicated a mismatch in emigration timing (Figure 8). The hatchery smolt wave comprised a shorter and more intense migratory period that was earlier than the natural-origin smolt wave. The downstream migration of hatchery-reared smolts stocked in the Piscataquis River suggests that hatchery smolts arrived at Brown's Mill Dam much earlier than did natural smolts (Figure 8). However, both natural-origin and hatchery smolt waves passed the dam while the smolt bypass was operational. The hatchery smolt wave originating in the main-stem Penobscot River arrived at Milford Dam earlier than the smolts stocked in the Piscataquis River because of their shorter migration distance and earlier release date (Figure 8). Most of the smolts stocked in the Piscataquis River (91.4%) and the main-stem Penobscot River (99.5%) passed Milford Dam outside of the supplemental spill period but not outside of the period when the downstream fishway was in operation. As noted previously, the natural-origin wave passed partly (14.6%) during the period when both protective measures (downstream fishway operation and supplemental spill) were in place at Milford Dam in 2019.

Details are in the caption following the image
Simulated 2019 migration waves for Piscataquis River natural-origin Atlantic Salmon smolts (light green) and multiple groups of hatchery smolts stocked at the Piscataquis River (dark green), on the main-stem Penobscot River upstream (yellow), and downstream (light yellow) of Milford Dam. The top panel illustrates the smolt waves passing Brown's Mill Dam. The middle panel illustrates the smolt waves passing Milford Dam, with the arrow indicating the timing of supplemental spill at that dam. The bottom panel illustrates the timing of all hatchery smolts stocked during that year (70,000) in addition to the natural-origin wave from the Piscataquis River (light green).

The majority (88%) of the 540,000 hatchery smolts were stocked approximately 1 km downstream of Milford Dam. Because 370,000 of these smolts were released before April 20 and the remaining 160,000 smolts were released between April 30 and May 3, there were two distinct peaks in their simulated arrival at the head of tide (14.5 km downstream of the stocking site). These analyses indicate that hatchery-stocked fish entered the estuarine phase of migration earlier than did natural-origin fish in 2019.


Based on more than two decades of trapping data, we demonstrated that although Atlantic Salmon smolt runs in Maine rivers are highly dynamic in timing, duration, and shape, the Tmean for the first quarter of the year explained a significant amount of the variation in the onset of emigration. The standardized smolt waves captured the long-term average run shape and duration for natural- and hatchery-origin smolts and thus can be used to set protection windows of sufficient duration to preserve adaptive variation in the population. Using the systemwide smolt migration model developed herein, we determined that the 14-d supplemental spill periods in place at Penobscot River dams during our study were too short to protect the peak of the smolt wave because the migration window can shift in time by more than 2 weeks and has an unpredictable distribution. Furthermore, the initiation temperatures for smolt emigration were more closely linked to the air temperature in the higher-elevation production areas (99–150-m elevation) where smolts originated than to the water temperatures recorded at Milford Dam (~30-m elevation), highlighting the importance of using local temperatures to set protective windows. Finally, the timing and migration dynamics differed between natural- and hatchery-origin smolts and indicated that integrating protective windows and stocking dates would help to protect smolts of both rearing origins. By basing the systemwide smolt migration model on standardized smolt waves and previously published estimates of smolt migration speeds (Stich et al. 2015a), this method has utility for salmon populations throughout Maine and can be used to inform smolt emigration dynamics for natural smolt production in Maine rivers that lack either current monitoring or a long time series. Additionally, the approach is transferable to other systems or regions with similar baseline data.

Characterizing Smolt Migration Timing and Patterns in Maine

The duration of annual smolt emigration in Maine typically exceeded 40 d, similar to other systems, where smolt runs have ranged in duration from 3 to 7 weeks (Whalen et al. 1999; Jensen et al. 2012; Thorstad et al. 2012; Harvey et al. 2020; Bjerck et al. 2021). The shape of the run differed among years (i.e., unimodal in some years and bimodal in others), likely due to interannual variability in temperature and discharge (Harvey et al. 2020). The standardized natural-origin wave that we developed accounted for annual stochasticity in run shape and duration by using emigration timing data from 20 years of trapping in the Narraguagus River. This median-centered averaging approach created a unimodal wave that is longer and lower in amplitude than the waves observed in unique years, but it serves as the best available wave for evaluation of protective measures at dams. It is essential to note that the duration and shape of the migration wave for a given population are a function of individual fish physiology, the location (within the watershed) and timing of the initiation of migration, diffusion, the migratory speed of individual fish, river temperatures, discharge, and the size and complexity of the watershed (Whalen et al. 1999; Stewart et al. 2006; Jensen et al. 2012; Michel et al. 2013; Harvey et al. 2020; Bjerck et al. 2021). Thus, efforts to refine individually based migration models or application of other modeling approaches (e.g., generalized additive models) may be useful next steps to facilitate prediction of more complex wave types (e.g., an early unimodal run versus an early bimodal run, as in Figure 4).

The standardized hatchery wave that was generated using the Narraguagus River trapping data set was unimodal, and the onset of migration occurred very rapidly after stocking—that is, hatchery smolts commonly initiated migration on the day of stocking. Rapid migration of hatchery fish has been observed previously in Maine (Stich et al. 2015a; Hawkes et al. 2017) and Norway (Urke et al. 2013). Individually based monitoring using telemetry in the Penobscot River demonstrated that the initiation of migration by hatchery fish averaged 2.5 d and rarely exceeded 1 week (Stich et al. 2015a). Our population-level data set was similar in that over 80% of smolts stocked in the Narraguagus River moved during the first week; however, our data set differed in that emigrants were captured as late as 23 d postrelease. This difference is likely due to the ability of trapping data sets to better document rare events with larger sample sizes than may be observed in telemetry studies with smaller sample sizes, although it could also be related to differences in the timing of stocking between the Narraguagus and Penobscot rivers.

During the years that we evaluated, the stocking of hatchery smolts in the Narraguagus River occurred later (early May of each year) than smolt stocking in the Penobscot River (mid to late April). However, stocking in the Penobscot River did occur in May during the years prior to our study. Although differences in release timing are important, there is little empirical evidence of extended stream residency under the temperatures typical of stocking in Maine rivers (Karppinen et al. 2014). In fact, smolts that were stocked into the Penobscot River early in the spring initiated their emigration more quickly than those stocked later in the season (Stich et al. 2015a). Therefore, the hatchery smolt emigration in the Narraguagus River appears to be typical of Maine watersheds of this size. We consider both the natural- and hatchery-origin standardized waves developed herein to be appropriate and informative because (1) the extensive time series in the Narraguagus River captured a wide variety of river conditions and thus emigration dynamics and (2) previously developed regional models have performed well for predicting smolt emigration dynamics over larger geographical areas (e.g., Vollset et al. 2021).

Predictive Run Timing Model

The Tmean for the first 90 d of the year was a strong predictor of the onset of smolt migration, explaining more than 60% of the variance in the DOY of 5, 25, and 50% smolt capture across four Maine rivers. Other work suggests that additional metrics, such as day length or discharge, are also contributing factors to the onset of smolt migration (McCormick et al. 1998; Stich et al. 2015a, 2015b; Simmons et al. 2021). Although multiple factors undoubtedly play roles in defining the onset of the run, our work and that of Vollset et al. (2021) indicate that air temperature for the first quarter of the year yields a robust annual predictor. Daily air temperature measured within smolt production areas of interest is available before the start of the spring smolt migration, making this a useful variable with which to annually predict the onset of smolt emigration and provide guidance to adaptively manage smolt protection windows (e.g., supplemental spill). This use of annual, local air temperature data also allowed us to account for different run initiation timing for each production area of interest within a given watershed, which subsequently affects when smolts encounter dams because of migration distance, speed, and juxtaposition (Zydlewski et al. 2005, 2014; Stich et al. 2015a, 2015b; Simmons et al. 2021).

The date of 25% capture in smolt traps thus provided a useful anchor point for describing the dynamics of a smolt run for the four rivers that were included in model parameterization. When applied to trapping data from the Pleasant and East Machias rivers, the regression model performed well, predicting the DOY of 25% capture within 5 d of the actual DOY in 9 of 11 cases. Our results concur with previous studies of smolt emigration dynamics. Indeed, the onset of smolt out-migration has classically been described as the point at which 25% of the run has passed collection stations used for documenting smolt runs (e.g., Otero et al. 2014; Vollset et al. 2021). However, it is inappropriate to use the date of 25% emigration as the trigger for implementing smolt protection measures at dams because conservation goals should include protection of population variability (Otero et al. 2014; USFWS and NMFS 2018; Thorstad et al. 2021).

Salmon conservation must strive to create protection windows that support and enhance the maintenance of temporal adaptive variation that is present in the population to increase survival probability and adaptation to climate change (Kennedy and Crozier 2010; Otero et al. 2014; Thorstad et al. 2021). Migration timing differs for smolts originating in different areas of the watershed, which results in smolts synchronizing their ocean entry and promotes local adaptation, with associated implications for reproductive success (Stewart et al. 2006; Garcia de Leaniz et al. 2007; Fraser et al. 2011; O'Toole et al. 2015; Mobley et al. 2019). Furthermore, some areas of a watershed will contain more thermal refuge habitat or have an overall cooler temperature regime, thus making them more resilient to climate change than other areas of the same watershed (Dugdale et al. 2015; Fullerton et al. 2015; Isaak et al. 2015; Mejia et al. 2020). It is imperative, therefore, that any protective measures encompass the full breadth of the run to conserve population complexity.

Setting Protective Windows to Conserve Population Variability

Our findings indicate that the operation of downstream fishways from April 1 to June 30, as is currently required at some Penobscot River dams, is adequate at present to encompass the entirety of the smolt run. However, continuing poor survival of smolts at dams necessitates the use of supplemental spill to reduce mortality due to the limited effectiveness of those downstream fishways (Brookfield Renewable Partners 2019). Supplemental spill is considered most important for improving smolt survival at dams when river discharge is low, but this measure comes at the expense of upstream fish passage and energy production during low-discharge periods. When river discharge is high, spill can occur regardless of prescribed supplemental spill windows, increasing the probability of smolt survival without impacting energy production. Indeed, Molina-Moctezuma et al. (2021) found that smolt survival at dams was greater during periods of greater discharge.

Under hydrological climate change scenarios, there is a considerable amount of variability expected with high- and low-flow conditions projected within the northeastern United States (Farr et al. 2021). However, rivers are predicted to warm faster and earlier in spring because of increasing winter air temperature and decreasing duration of ice cover (Hayhoe et al. 2007; Sharma et al. 2019; Gillis et al., in press). In New England rivers, winter air temperature was significantly correlated with the timing of snowmelt-driven streamflow (Hodgkins et al. 2003). Snowmelt-driven streamflow (measured as WSCVD) in the eastern United States has advanced by more than 8 d since 1940, largely due to warmer air temperature (Dudley et al. 2017). As a result, peak streamflow is occurring earlier in the smolt emigration period. Peak streamflow occurring out of phase with peak out-migration is expected to increase the need for supplemental spill to ensure adequate protections of emigrating smolts at dams in Maine. Our systemwide migration model can help to proactively define the supplemental spill window more precisely, thereby optimizing the protection of smolts while also accounting for the interannual variability in streamflow that is increasing due to climate change (Hodgkins and Dudley 2006).

Application of the systemwide smolt migration model to evaluate protective measures in place for Penobscot River dams during 2021 highlighted that when protective measures are too short in duration or are based on inappropriate triggers (i.e., river temperature measured at a low-elevation dam), even the core of the run may pass dams outside of protection windows. Our estimates indicate this to be the case for both Milford and West Enfield dams, as almost no smolts passed during the supplemental spill period. A substantially greater percentage of the run was estimated to have passed Mattaceunk Dam during the supplemental spill period due to the longer spill period at that dam, which was triggered by temperature (10°C) measured in proximity to the high-elevation production area.

We have demonstrated that the systemwide smolt migration model is a useful tool for predicting the onset of smolt out-migration. However, as noted previously, the use of median-centered catch data means that the smolt migration model will predict the emigration timing for a unimodal smolt wave only. Therefore, use of the smolt migration model could still result in protective measures (e.g., supplemental spill periods) that are out of sync with the actual smolt run if those measures are too short in duration, particularly during years with more complex wave types. We designed the systemwide smolt migration model to be of sufficient duration to protect the tails of the run; for Maine rivers, this duration was identified as 54 d. Attempting to set shorter protection windows using the systemwide smolt migration model could result in measures that are ineffective at protecting smolts if river conditions (e.g., temperature and discharge) result in a smolt run that becomes either accelerated or delayed. Given that the maximum difference between observed and predicted DOYs of 25% smolt capture was 9 d, broadening the windows to start 9 d earlier and end 9 d later would result in a protective window of 72 d, providing a buffer for variability in the predicted onset of run timing.

Supplemental spill and bypass facilities are not the only mechanisms at dams that influence survival. As noted previously, increased river discharge during wetter springs can increase smolt survival during passage through dams (Molina-Moctezuma et al. 2021). Consequently, passing a dam outside of smolt protection windows does not mean certain mortality; likewise, passing a dam during a protection window does not ensure survival. Furthermore, the Atlantic Salmon is only one of a dozen migratory species in Maine rivers and additional work is needed to support restoration of the entire migratory fish community (Hare et al. 2019). Although actions taken for salmon may be helpful for other species, differences in swimming ability and run timing may require alternative measures. Physical design, like screens on turbine intakes, can reduce mortality from turbine entrainment (NMFS 2011; Thorstad et al. 2012). If permanently installed, physical designs would not only yield benefits for smolts, but also provide protection for other emigrating diadromous species and could even eliminate the need for extended periods of supplemental spill.

Optimizing Hatchery Smolt Stocking

Our results also suggest that the stocking of hatchery smolts is asynchronous with both the emigration timing of natural-origin smolts and current protective windows. This asynchrony could impact not only the effectiveness of protective measures for the overall population, but also survival during saltwater transition and the marine phase (Hvidsten and Johnsen 1993). Ocean entry timing needs to occur when smolts are physiologically capable of tolerating the transition to salt water (McCormick et al. 1998; Stich et al. 2015b). Furthermore, salmon ocean entry is generally thought to be timed to optimize early ocean feeding and is likely a strong selective factor in a changing climate (Kallio-Nyberg et al. 1999; Otero et al. 2014; Hawkes et al. 2017; Thorstad et al. 2021).

Application of the systemwide migration model could be used to (1) ensure that smolt protection measures are in place during the full period when hatchery-origin smolts are expected to be migrating through dams; and (2) fine-tune the stocking dates for hatchery smolts to reduce the mismatch in migration timing with natural-origin smolts, which may optimize survival, both in river and upon ocean entry. Karppinen et al. (2014) demonstrated that the survival of hatchery-origin Atlantic Salmon smolts through the regulated River Oulujoki (Finland) was greater when stocking occurred at the optimal temperatures for migration (8–10°C) and concurrent with the run timing of natural-origin smolts. Furthermore, returns of hatchery-origin adult Atlantic Salmon to the River Orkla (Norway) were greater when stocking was synchronized with large groups of migrating natural-origin smolts than when hatchery fish were released with smaller numbers of natural-origin smolts (Hvidsten and Johnsen 1993).

Stich et al. (2015a) evaluated eight Penobscot River hatchery cohorts and found that all release groups experienced thermal conditions below thresholds for the loss of smolt characteristics and that most smolts were released before or near the physiological peak (Handeland et al. 2004). This suggests that a relatively broad window of release dates could be optimized without compromising physiological readiness. Stich et al. (2015a) proposed a knowledge-based approach to guide the timing of stocking by using discharge, river temperature, distance from the ocean, and hatchery ATUs to optimize the survival of hatchery smolts. Integrating their approach with our smolt migration model and projected discharge data from the NOAA Advanced Hydrological Prediction Service would enable managers to plan optimal release dates for hatchery smolts by early April of each year. In this way, hatchery stocking locations and timing can be aligned with protective measures at dams to protect the maximum numbers of both natural- and hatchery-origin fish and to enable best practices for stocking smolts into the higher-quality habitat upstream of dams (Stevens et al. 2019).


The recovery strategy for the federally protected GOM DPS of Atlantic Salmon focuses on the freshwater and estuarine phases of the life cycle, and a key strategy is increasing the production and survival of smolts (USFWS and NMFS 2018; USASAC 2021). However, direct and latent mortality of salmon smolts at dams substantially reduces the numbers of smolts entering the ocean (Stich et al. 2015a, 2015b; Stevens et al. 2019). Therefore, it is imperative for all protective measures at dams (e.g., supplemental spill or operation of bypass facilities) to occur at appropriate times and to be of sufficient duration to maximize smolt survival. We determined that the 14-d supplemental spill window triggered by attaining a threshold of 10°C was inadequate to protect even the peak of the smolt run in most years because of interannual variation in the timing and shape of the smolt run. Adoption of the systemwide migration model could help decision makers to refine the timing of smolt protection measures on an annual basis. Because the model was developed from data sources available by early April, this approach allows flexibility that is both proactive and responsive to interannual environmental variability. Use of annually available air temperature data collected within smolt production areas allows managers to account for differential run timing and migration distance experienced by smolts from different areas of larger watersheds. In this way, the smolt wave model (1) accounts for the fact that the smolt run comprises multiple populations of smolts with different run timing and (2) allows managers to consider the effects that the production area of origin (and associated thermal exposure and migration distance) and the production source (natural or hatchery production) have on arrival timing at dams or ocean entry. Our results also set the stage for reducing the mismatch in timing between hatchery- and natural-origin smolt migration and aligning spill with stocking dates to better protect hatchery smolts that are released upstream of dams. Finally, such a holistic view toward developing smolt protection measures must integrate the timing and duration components described herein with the continued monitoring of survival.


We thank Robert Dudley (USGS) for helping with provisional WSV and WSCVD calculations, Matt Dzaugis (Gulf of Maine Research Institute) for extracting the North American Regional Reanalysis temperature data for all smolt trapping locations, and Sean Birkel (University of Maine Climate Change Institute) for providing us with the measured temperature data for all production areas. Paul Christman (Maine Department of Marine Resources) and Justin Stevens (Maine Sea Grant) documented river reach lengths. We thank Dan Tierney (NOAA Greater Atlantic Regional Fisheries Office) for creating Figure 1. Jeff Murphy, Matt Buyhoff, and Dan Tierney (NOAA Greater Atlantic Regional Fisheries Office); Casey Clark and Sean Ledwin (Maine Department of Marine Resources); and Joe Zydlewski (USGS Cooperative Research Unit) provided invaluable information on the smolt protection measures in the Penobscot and Kennebec rivers as well as feedback and discussion that improved our analyses. Steve Gephardt, Benjamin Burford, and an anonymous reviewer provided comments that greatly improved the manuscript. Finally, we thank the many biologists, technicians, and partners who conducted the daily smolt trapping activities, which produced the data sets that were integral to our analysis; we especially appreciate Ernie Atkinson, Colby Bruchs, Paul Christman, Graham Goulette, Christine Lipsky, Jennifer Noll, Peter Ruksznis, Mitch Simpson, Randy Spenser, and staff at the U.S. Fish and Wildlife Service's Green Lake and Craig Brook National Fish Hatcheries.