Bigmouth Buffalo (Ictiobus cyprinellus) migratory behaviour and seasonal home range overlap are functions of geographic space in a fragmented riverscape

The study area is contained within the Lake Winnipeg watershed (∼1 000 000 km²), which encompasses parts of Alberta, Saskatchewan, Manitoba, Ontario, Minnesota, North Dakota, South Dakota, and Montana (Fig. 1—inset map). Land use in the watershed is dominated by cropland agriculture with considerable nutrient inputs draining north into Lake Winnipeg (ECCC 2020). In the current study, 80 Bigmouth Buffalo were tracked in the Red River sub-watershed, including the eastward flowing Assiniboine River and northward flowing Red River of the North (herein referred to as the Red River; Fig. 1). Given the low elevation landscape, the Red River sub-watershed is prone to both recurring flooding and periods of low discharge.

The Red River is the largest river in the sub-watershed and forms the boundary between Minnesota and North Dakota, flowing approximately 885 river kilometers (rkm) north from Minnesota/North Dakota (ND), USA to Manitoba (MB), Canada. It is a low-energy, suspended sediment, mud-dominated, meandering river that occupies a shallow valley eroded into a clay plain with an alluvium that is primarily silt (Brooks 2003; Brooks 2017). Within the United States the river is typically one continuous run with some riffles present in the sections between Fargo and Riverside (F-R) and Wahpeton to Fargo (W-F; Fig. 1). Between Wahpeton and Drayton (W-F, F-R, R-D; Fig. 1), the river has high sinuosity (ranges from 1.7 to 2.3) and a low gradient (ranges from 0.04 to 0.25 m·km⁻¹; Topp et al. 1994). In Canada (D-S, S-L, LW; Fig. 1), the Red River widens to eventually form a large delta (Netley-Libau Marsh) at the south end of Lake Winnipeg. Most of this reach is influenced by Lake Winnipeg seiche.

There are four types of barriers (lock and dam, low-head dam, radial arm floodgates, rock arch rapid (RAR)) that may cause fragmentation along the river, including: Kidder RAR (ND), Hickson RAR (ND), Christine RAR (ND), Fargo South RAR (ND), Midtown RAR (ND), Fargo North RAR (ND), Riverside RAR (ND), Drayton Dam (ND), Red River Floodway Control Structure (MB), and St. Andrews Lock and Dam (SALD; MB) (Fig. 1). We evaluated migration as well as summer and winter home ranges within and across sections of the Red River that were divided by partial movement barriers (Fargo North RAR, Riverside RAR, Drayton Dam, SALD; Fig. 1). These structures were previously shown to restrict movement resulting in low transition probabilities (Enders et al. 2019). Further details of these sections and partial movement barriers are described below.

The cities of Fargo-Moorhead along the Red River were the locations of the Fargo North (built in ca. 1933), Fargo Midtown (built in ca. 1961) and Fargo South (built in ca. 1933) dams, originally constructed for flood control. From 1999 to 2003, these three dams were converted to RAR to eliminate a hydraulic roller drowning hazard, mimic the microhabitat of natural rapids, and improve fish passage (Aadland et al. 2005). The resulting structures provide a step-pool channel with a centerline slope of 5% to allow passage of weaker swimming species (Aadland 2010).

The Riverside Dam (built ca. 1922) was located in the City of Grand Forks, North Dakota, approximately 476 rkm from the mouth of Lake Winnipeg. Originally designed for municipal water supply, the structure was modified in 2001 into one of the world's largest full river width RAR to manage erosion, eliminate a dangerous hydraulic roller, and provide fish passage and spawning habitat (Aadland 2010). The conversion from low-head dam to rapids was achieved by constructing a wedge of fieldstone and cobble into a 5% slope (3% near shore) over the existing dam, above which ten boulder weirs now provide grade control, pool habitat, and fish-passable velocities (Aadland 2010).

The Drayton Dam at rkm 327 was constructed in 1964 to supply water for agricultural and municipal uses. The dam was originally designed as a concrete weir with a spillway length of 68.5 m and a crest elevation of 3.7 m above the natural riverbed. Operated as a run-of-the-river water control structure, the Drayton Dam was never intended to readily facilitate fish passage. The most recent evaluation of fish movement indicates that upstream passage is indeed limited for Bigmouth Buffalo (Enders et al. 2019).

The Red River Floodway, a trapezoidal-shaped diversion canal about 48 km long and 160–370 m wide, was constructed in response to a devastating flood in 1950 (Brooks 2017). The floodway begins just upstream (south) of Winnipeg, bypassing the city on the east side, before rejoining the river just downstream of the St. Andrews Lock and Dam. The Floodway Control Structure is just downstream of the Floodway entrance in the mainstem of the Red River. It regulates the Red River through Winnipeg when discharge exceeds 900–1000 m³·s⁻¹ by raising submerged radial arm gates to divert a portion of flow into the Floodway channel (Brooks 2017). During operation, the control structure is a barrier to upstream fish movement.

The St. Andrews Lock and Dam (SALD) at rkm 44 is located in Lockport, Manitoba. The lock and dam was operational in 1910 and designed for flood control and navigation over Lister Rapids. In 1913, the structure was retrofitted with a concrete alternate orifice and slot fishway with two sloping chutes running north and south. Fish can move upstream through the fishway or lock, when operated, and downstream through the fishway, lock, or over the dam (Enders et al. 2019). Species found in the fishway include Walleye (Sander vitreum), Sauger (Sander canadense), Saugeye (S. vitreum x S. canadense), White Sucker (Catostomus commersonii), Shorthead Redhorse (Moxostoma macrolepidotum), Channel Catfish (Ictalurus punctatus), Common Carp (Cyprinus carpio), Silver Chub (Macrhybopsis storeriana), River Shiner (Notropis blennius), Mooneye (Hiodon tergisus), Goldeye (Hiodon alosoides), White Bass (Morone chrysops), and Freshwater Drum (Aplodinotus grunniens) (Willis 1994). Although Bigmouth Buffalo have not been documented in the fishway, they are known to pass the St. Andrews Lock and Dam, both upstream and downstream (Enders et al. 2019).

Acoustic telemetry receiver stations (Innovasea, VR2W and VR2Tx, Nova Scotia, Canada, n = 235) were established in the Red River, Assiniboine River, and Lake Winnipeg beginning in 2016. Because the current study is part of a larger, multi-species (n = 9) fish movement research and monitoring program in the basin, receiver distribution and spacing was decided by two main factors: (1) equipment availability through time and (2) location and number tagged. Receivers were gradually added to the study system as they became available (Table S1). Most fish in the overall program have been tagged in S-L, LW, and near the city of Winnipeg, with relatively less tagging outside of Canada. Based on these factors, receiver spacing in the Red River was 30 km in the USA portion of the river, 20 km from the USA border to the south end of Winnipeg, 10 km from Winnipeg to the St. Andrews Lock and Dam, and 5 km from St. Andrews Lock and Dam to Lake Winnipeg (Fig. 1). Five receivers were deployed with a 30 km spacing in the Assiniboine River up to the Portage Diversion (rkm 151; Fig. 1). In 2016, receivers deployed at Drayton, ND, Halstad, MN, and Grand Forks, ND were replaced with the 30 km spacing design. Receivers were further deployed in distributary channels and wetlands associated with Netley-Libau Marsh (n = 8) and into several smaller Red River tributaries, including: Devil’s Creek, Cook’s Creek, Seine River, La Salle River, Rat River, Roseau River, Red Lake River, and Sheyenne River. Receivers in Lake Winnipeg were arranged in a 5 km grid from the mouth of the Red River to just north of Gimli, a 7 km grid from Gimli to the Narrows, and a 14 km grid into the North Basin. By 2022, the total distance covered by the receiver network in rivers was approximately 860 rkm (Fig. 1). All river deployments were switched to VR2Tx receivers in 2017, with the exception of section LW where ∼30% of receivers were still VR2W models in the final study year (Fig. 1, Table S1). Range testing was conducted at a receiver station in the Red River (rkm 24) in 2016, 2017, and 2018 (Table S2).

Receivers were connected to a 12.7 mm threaded rod (∼50 cm long) with four cable ties. The threaded rod was fixed to the center of a ∼45 cm × 45 cm × 10 cm granite block that weight ∼50 kg. A rope ∼12 m long was attached to an eye bolt fastened to the top of the threaded rod. The rope was further attached to a ∼11-14 kg cross net anchor. The anchor was deployed upstream at which time the boat drifted until the rope was nearly tight. The granite block was then lowered in an upright orientation to the bottom, near the thalweg.

Bigmouth Buffalo (n = 80) were captured by boat electrofishing at eight tagging locations in the Red River drainage (Fig. 1). Tagging efforts spanned 2 years with 40 fish tagged in the La Salle (n = 20) and Seine Rivers (n = 20) in 2016. The additional 40 fish were tagged in 2017 between the Fargo/Sheyenne (n = 8), Halstad (n = 4), Drayton (n = 8), St. Andrews (n = 9), Selkirk (n = 9), and Netley-Libau Marsh (n = 2) tagging locations. Tagging efforts targeted Bigmouth Buffalo near the confluence of major tributaries, downstream of dams, and floodplain habitats. All Bigmouth Buffalo were tagged within 4 rkm of the Red River proper (Fig. 1). Fish were indiscriminately captured and held in tanks filled with ambient river water before being measured for weight (nearest gram), fork length (mm), and total length (mm). Tag burden was minimized by only retaining animals with a body mass > 1.2 kg (<2% tag:body mass; Crossin et al. 2017). To avoid the use of chemical anesthesia, retained fish were placed into the Portable Electroanesthesia System (PES., Smith-Root, Vancouver, WA, USA) for immobilization during surgery. The PES was set to 100 Hz, 25% duty cycle, and 40 V. Pulsed direct current is an appropriate sedation for adult fish because it provides a surgery window of 250–350 s and fish recover quickly with minimal impact to vertebral integrity (Vandergoot et al. 2011). Upon sedation, fish were placed in a padded v-shaped trough. Ambient river water was continuously pumped over the gills using a recirculating flow-through pump system to maintain normal respiration during the surgical period (<5 min). A small incision was made posterior to the pectoral girdle just dorsal of the ventral midline. Sex was determined by attempting to examine gonads through the incision; however, the sex assignment was uncertain for >50% of tagged fish. An acoustic transmitter (Innovasea, V16-4 H, 16 mm diameter, 24 g, 6.5-year battery life, average transmission delay of 120 s with a pseudo-random uniform interval between 80 and 160 s) was inserted posteriorly into the peritoneal cavity. The incision was closed with three to four interrupted stitches (standard surgical knots; 3-0 polydioxanone-II violet monofilament suture; Ethicon, Cincinnati, OH, USA). The tag and all other surgical equipment were soaked in a disinfectant (Betadine; Avrio Health L.P., New York, New York) and rinsed in filtered water prior to use. Surgical gloves were worn to reduce the risk of infection. Post-surgery, fish were placed in a recovery tank for 10–15 min. All fish achieved full recovery as evidenced by their ability to maintain equilibrium and strong reflex reaction to grabbing the caudal peduncle. Upon recovery, fish were released at the tagging location. Fish capture and surgical procedures were approved by Fisheries and Oceans Canada's Ontario, Prairie and Arctic Animal Care Committee (OPA-ACC; FWI-ACC-2016-018, FWI-ACC-2017-001), following the Canadian Council on Animal Care (CCAC) guidelines; and the University of Nebraska-Lincoln's Institutional Animal Care and Use Committee (IACUC; Project ID: 1208), accredited by the Association for Assessment and Accreditation of Laboratory Animal Care (AAALAC).

We used a one-way ANOVA to test for differences in body size (fork length) among tagging locations. Length was log-transformed to achieve homogeneity of variance as indicated by a Levene Test. Pairwise comparisons were evaluated with a Tukey HSD test (α = 0.05). Residuals were visually inspected to further validate the model (Zuur et al. 2010).

Environmental variables for the Red River were downloaded from online gauging datasets using R packages tidyhydat (Water Survey of Canada, Albers 2017) and dataRetrieval (USGS, De Cicco et al. 2022). Starting in 2017, most receiver models in the river network were VR2Tx that also recorded temperature. Water quality data, collected bi-weekly during the open water season by the City of Winnipeg for each year (e.g., https://legacy.winnipeg.ca/waterandwaste/sewage/monitoring/2017RiversReports.stm), were included to increase the spatial resolution of turbidity and water temperature data. Temperature, discharge, and turbidity data (Fig. 2) were interpolated for any given location and date using distance-based interpolation from the gstat package in R (Pebesma 2004; Gräler et al. 2016).

Erroneous detections (e.g., due to noise or code collisions) and impossible movements were removed prior to the analysis (Pincock et al. 2010). Initial data processing was completed using the R package actel, which provides a reproducible method for identifying erroneous detections, generating outputs for future analysis, and visualizing fish movements (Flávio and Baktoft 2020). Actel computes movement events based on thresholds of time between detections and minimum number of detections. The initial data filtering procedure identified movements for individual fish when at least two detections were recorded on a receiver. A new movement event was logged when the fish was detected again outside of the one-hour time interval at any given receiver (Flávio and Baktoft 2020).

Of 3 689 629 detections, 430 were identified as erroneous based on speed and jump checks in the actel data filtering system. These erroneous detections arose due to overlapping detection ranges and impossible movement speeds among receivers in the Netley-Libau Marsh and the Winnipeg River. Of the 80 tagged Bigmouth Buffalo, one fish was never detected, and nine others were not detected for longer than 1-year post-release, with many failing to transmit in the final year of monitoring (2022) due to battery depletion. Only fish that were considered active for the entire year (detected before the first day of the year of interest and detected after the year of interest) were used for migratory analysis (see below). This criterion subsequently removed the movements from the year when a fish was tagged. The number of individuals included in the analysis ranged between 36 and 69 each study year from 2017 to 2022 (Table S3).

We defined migration as the directed movement of an individual over a distance greater than that associated with regular daily activities, resulting in redistribution across the landscape for some fitness benefit (Dingle and Drake 2007; Semlitsch 2010). Under this definition, we classified the movements of Bigmouth Buffalo through visual inspection of latitudinal positions over time, changepoint analysis, and metrics that are associated with migratory behaviour (successive unidirectional movements at increased speed). Changepoint analysis allows for the identification of points in a time series where a statistical property changes rapidly (Killick et al. 2012). Due to the orientation of the Red River and the direction of migratory movements, latitude was a sufficient proxy for a fishes broad-scale position in the riverscape (Fig. 1). Detections in the Assiniboine River were excluded from migration assignment because of the small number of fish (n = 4) that occupied the river for extended periods of time. We converted the detected locations of each Bigmouth Buffalo into a time series of daily latitude using the last observation carried forward (LOCF) to fill in days where the fish was not detected (Colborne et al. 2019). The daily latitude of each individual was subset by year before changepoints were identified based on the mean using the changepoint package in R (Killick and Eckley 2014; Killick et al. 2022) and the binary segmentation method, specifying a maximum number of two changepoints (one fall and one spring migration in a year). Outside of the previously documented periods for migrations (April—May, September—October), Bigmouth Buffalo home ranges are typically less than 90 rkm (Enders et al. 2019). Therefore, if an individual made successive unidirectional movements—at speeds > 0.5 m·s⁻¹—to a site > 90 rkm away this increased our confidence that the movement was associated with a migration. Movement speed was calculated automatically by the residency analysis in actel.

Migration assignment was based on a set of rules (see Box S1 for detailed definitions) and conducted by two researchers independently. Discrepancies in the assignments were resolved by joint consensus. Upstream migration was assigned to southward movements from a summer to a winter site, and downstream migrations were assigned to northward movements from a winter to a summer site. We deemed the end of a migration as the first time of arrival on the last day of directed movement (i.e., subsequent movement event was at the same receiver) or the time before a movement in the opposite direction. If an individual appeared to stop or change direction, but then continued its initial path within a two-week period, the movement was still considered to be a migration.

Once each migratory event (upstream departure, upstream destination, downstream departure, downstream destination) was classified, we summarized the characteristics of these events by determining Julian day, water temperature (°C), and latitude of each event. We also determined the total migration distance (rkm) for the upstream and downstream migrations. The cumulative number of fish migrating each week was overlayed on a heatmap of interpolated temperature, flow, and turbidity. To define the migration periods, yearly logistic curves were fit using nonlinear least squares to the cumulative number of fish that migrated each week. A logistic curve has three parameters, the asymptote (aymp = 1), a midpoint when rate is steepest (mid), and a scale parameter which sets the slope of the curve (scale). The logistic curves were fit with nonlinear least squares using a self-starting logistic model (SSLogis) from the R package stats (R Core Team 2023). To determine the statistical significance of the coefficient estimates, we calculated the p-value associated with the Wald statistic. The fitted curve was then used to determine the spring and fall migration periods as all values >0 and <1 for the model fit to the cumulative number of migrators per week in the respective period.

Observations of migratory events for individuals on multiple occasions allowed us to calculate repeatability or the amount of behavioural variation arising from individual differences (Bell et al. 2009). Here, repeatability can be considered as the proportion of total variation in migration that can be attributed to differences between (compared to within) individuals (Nakagawa and Schielzeth 2010). We determined the repeatability and adjusted repeatability of characteristics (i.e., Julian day, water temperature, and latitude of departure and destination; and distance moved) associated with each Bigmouth Buffalo migration (herein referred to as migratory event characteristic) using linear mixed effects models for each migratory event characteristic as the response variables, and transmitter as a random effect (Nakagawa and Schielzeth 2010). Repeatability and adjusted repeatability were calculated using the rptR package in R calculating 95% confidence intervals from 10 000 bootstrap iterations and p-values were estimated using likelihood ratio tests, with a significance level of α = 0.05 (Stoffel et al. 2017). Year was tested as a fixed effect to account for: (1) annual changes in confounding factors that might influence phenology of the trait in question (Biro and Stamps 2015); and (2) estimate relative repeatability across years (i.e., adjusted repeatability), which is typically greater than the estimate of repeatability that ignores time related change (Kürten et al. 2022). Statistical assumptions were visually assessed and validated for each model.

Daily LOCF datasets of fish locations were split into winter and summer seasons for all fish—regardless of meeting the conditions for migration—by removing the period defined as spring or fall migration using the fitted values from the logistic regression. Overlap between summer and winter ranges was determined using the R package riverdist (Tyers 2022), where values ranged from 0 to 1. A value of 0 represented no overlap whereas all values > 0 were converted to 1 and represented overlap between summer and winter ranges. River sections were defined by potential impediments to movement (Fig. 1). These structures were previously shown to restrict movement resulting in low transition probabilities (Enders et al. 2019). Fish were assigned to the river section in which they cumulatively spent the most time per season. This assignment was then compared to where they spent the most time overall to determine site fidelity (i.e., a binary indicator of whether the fish were in the area most used). Movements into tributaries were also included in seasonal linear home range, with tributaries included in the section where their confluence occurred. Fish that spent most of their time in the Wahpeton to Fargo (W-F) or Lake Winnipeg (LW) sections for a given season were not included in the analysis due to low sample size upstream of the Fargo North RAR and the change from linear (upstream or downstream) to multidirectional movement paths associated with the lake. The proportion of the section used was calculated by dividing home range size by total length of the most used section; values greater than one (i.e., indicating passage across at least one of the obstacles dividing sections) were set to one. Linear mixed models (lme4, Bates et al. 2015) were used to determine: (1) the effect of river section, section use, year, and season, grouped by individual (random intercept), on log-transformed seasonal home range size (distribution = Gaussian); and (2) the effect of river section, section use, grouped by individual (random intercept), on the probability of overlap (distribution = binomial, link = logit). The model was validated by visual inspection of the residuals. While home range and body size are typically related (Mcloughlin and Ferguson 2000), Bigmouth Buffalo lengths were considerably skewed among river sections, thus preventing us from including this term in the analysis of home range size and overlap.

Bigmouth Buffalo only met the criteria for migration in the longest unimpeded section of river (Fig. 3). Generally, these movements were directed downstream in the spring, departing from areas near Drayton Dam to locations around the La Salle and Seine rivers in Canada where a large proportion of fish were tagged (mean distance = 210 rkm ± 107 SD). Subsequent upstream movements in the fall were from Canada to locations in the United States, generally downstream of Drayton Dam (rkm 280–310; 82% of destinations). In smaller sections of the river and the Assiniboine, movement patterns tended to be less directed (Fig. 3). Individuals in the section downstream of SALD moved regularly between the dam and into Netley-Libau Marsh, occasionally spending time in Lake Winnipeg. One individual moved approximately 60 km from the mouth of the Red River to Traverse Bay in the southeast portion of Lake Winnipeg. Individuals in the upper portions of the array (sections R-D, F-R, W-F; Fig. 1) often remained in their tagging section, but some individuals made longer distance movements outside of the reach in which they were tagged. Bigmouth Buffalo movement patterns were typically within the Red River proper for the majority of this study. Only two individuals spent greater than 50% of their time in tributaries, but over 88% were detected moving into a tributary at some point. The average proportion of time spent in tributaries was estimated at 13.2% (SD = 16.3%).

Migratory individuals generally moved from the United States to Manitoba in the river section between SALD and Drayton Dam. Spring movements from overwinter to spawning/summer ranges often began between rkm 252 and rkm 370 (near Drayton Dam; 70% of departures). Directed movements typically ceased between rkm 102 and rkm 112 (near the La Salle River; 63% of destinations). Most downstream migratory movements occurred after ice out when temperatures were rising. Downstream departure water temperatures averaged 10.6 °C (SD = 5.92), opposed to destination temperatures on the day of arrival of 12.6 °C (SD = 5.48). Upstream migration movements in the fall usually started between rkm 72 and rkm 132 (near the city of Winnipeg; 81% of departures) and ended below the Drayton Dam (rkm 280–310; 82% of destinations). The notable exception to this pattern occurred in 2019 and 2022, the only years that directed movements concluded upstream of Drayton Dam. Upstream migrations coincided with decreasing temperatures where departures (16.5 °C ± 0.7 SD) had higher temperatures than the destination areas (12.9 °C ± 1.0 SD) on the day of arrival. Temperature on the day of arrival to downstream destinations was on average 3.6 °C lower than the temperature on the day of departure (paired one way t test,p < 0.05).

The average speed of fish migrating upstream was 0.273 m·s⁻¹ (n = 1093, SD = 0.141) with a range of 0.003–0.883 m·s⁻¹ compared to fish migrating downstream with an average of 0.715 m·s⁻¹ (n = 988, SD = 0.464) and range of 0.002–1.960 m·s⁻¹. Speeds during non-migratory behaviours averaged 0.374 m·s⁻¹ (SD = 0.323) with observed maximum speed of 2.46 m·s⁻¹. Movement speeds exceeding 1.75 m·s⁻¹ (n = 20) were only observed between receiver arrays with spacing of less than 10 rkm. Movement speed had a weak positive correlation with temperature (r = 0.21, p < 0.01) and day length (r = 0.30, p < 0.01).

There were differences in the proportion of migratory individuals between years and between tagging locations (Fig. 4). Fish tagged in the La Salle and Seine rivers consistently had higher proportions of migratory individuals than the other tagging locations. There was additional variation in the proportion of migratory individuals across years within tagging groups; however, the proportion of migratory individuals in the entire tagged population remained relatively stable (Fig. 4). Of Bigmouth Buffalo that migrated at some point (n = 39), most individuals (n = 31) skipped migratory movements during at least two years of observation. Of individuals that were consistent in migratory behaviour (skipped one or fewer migrations, n = 8), all except one of those individuals were tagged in the La Salle or Seine rivers.

Most (75%) of the 20 fish tagged downstream of SALD were able to pass upstream of the structure at some point and three individuals passed three or more times. Upstream passages only occurred between April and September, with the highest number recorded in August (n = 11). Seven fish that successfully passed SALD also made migratory movements. Only one of these individuals showed consistent annual migratory behaviour after passing the dam. While eight of the fish did not migrate after passing SALD, most remained in upstream sections for several years. For example, after passing SALD in the fall of 2019, one individual swam to areas near the Halstad tagging location where it remained for the entire study (Fig. 1). Bigmouth Buffalo were not observed traveling upstream through the Red River floodway control structure while it was in operation (Fig. 1).

The majority of tagged individuals were able to pass upstream over Drayton Dam during the study (n = 43), with a total of 120 upstream passage events. Nine fish were detected at the closest downstream receiver but never passed the dam. Due to receiver spacing, we cannot determine if those fish made passage attempts. Upstream passages over Drayton Dam did not occur outside of March-October. April (n = 48) and May (n = 29) had the highest number of passes followed by September (n = 12), October (n = 11), June (n = 11), July (n = 5), March (n = 2), and August (n = 2). Many of the spring passes were associated with migratory individuals that overwintered downstream of the dam and moved upstream before making a return trip downstream. Upstream passage over the other dams was less common, with 27 of these events recorded at the Riverside RAR and only four over the Fargo RAR.

Repeatability analyses for Bigmouth Buffalo phenology of migratory events showed that individual repeatability was <0.4, such that fish reached their winter destinations with some annual consistency (Julian day; upstream destination: R = 0.192, p = 0.023; Table 1). Further, individuals tended to depart from the same winter location (latitude of upstream departure: R = 0.215, p < 0.015; Table 1) and returned to similar spawning/summer locations (latitude of downstream destination; R = 0.548, p < 0.001; Table 1; and downstream departure: R = 0.222, p < 0.008; Table 1) annually. Stream temperatures at the onset of downstream migration were consistent across years (temperature of downstream departure: R = 0.328, p = 0.002; Table 1). The timing of spring migration (Julian day of downstream departure), the distance of fall migration (distance upstream), the temperature when spring migrations ended (downstream destination temperature), and the latitude of winter locations (latitude of the upstream destination) were inconsistent across years (p > 0.05 in all instances; Table 1). Adding year as a fixed effect to the linear mixed effect models prior to calculating repeatability accounted for some of the model variance and increased estimates (Table 1).

The start, end, and length of the migration periods were variable across years and seasons (Fig. 5). Spring migration was initiated alongside distinct changes in several measured environmental variables including an increase of water temperature, higher flows, and a spike in turbidity (Fig. 5). No clear change was observed in the measured environmental variables to indicate the end of spring migration or start of fall migration (Fig. 5). The midpoint of the spring migration ranged from early April (week 14 in 2021) to early May (week 19 in 2020), depending on year (Table 2). The widest spring migration start date range was in 2022 (scale parameter = 0.67) and the narrowest in 2021 (scale parameter = 2.70, Table 2). Fall migration midpoint ranged from late September to late October (Table 2). The widest fall migration start date range was in 2018 (scale parameter = 0.61) and the narrowest in 2021 (scale parameter = 1.68, Table 2).

On average, winter mean home ranges were smaller (23.5 km) than summer mean home ranges (48.9 km, Fig. 6a). On average, seasonal home ranges encompassed a greater percentage of the total river section in summer (35.4%) than winter (13.3%, Fig. 6b) and was highest in summer in the section of river between SALD and Lake Winnipeg where measured geographic space was most limited (90%, Fig. 6b). The top two models for summer and winter home range size included season, and one of the top two models further contained site fidelity (top model is presented in Table 3), all other models ΔAIC > 2 (Table S4). Home range size was significantly larger in the summer and when fish were outside their most used section. The proportion of variance explained by the random effect of individual ID was low (ICC = 0.06) and including ID as a random effect was a small improvement to the explained variance (marginal R² = 0.128, conditional R² = 0.184). There remained a sizable portion of the variance in home range size not explained by the model. Importantly, the absence of river section from the top model indicated that it provided little explanatory power (Table 3).

In most instances, there was no overlap between the winter and summer ranges (198/321 observations). The average winter and summer range overlap was 26.8% when overlap occurred (overlap > 0; Fig. 6c). The top two models for summer and winter overlap included river section, and one of the top two models further contained site fidelity, although site fidelity was not significant (top model is presented in Table 3), all other models ΔAIC > 2 (Table S4). The probability of overlap was highest downstream of SALD (Table 3, Fig. 6d). Although higher than the home range model, variance explained by including individual ID as random effect was low (ICC = 0.21). Including ID as a random effect more than doubled the variance explained (marginal R² = 0.134, conditional R² = 0.315).