Study Area

Data were collected within Ottawa National Wildlife Refuge in Oak Harbor, Ohio, USA (41° 37' 22.6”N, 83°12' 37.5”W). This federal conservation area is managed by the U.S. Fish and Wildlife Service (USFWS) and located on the southwest shore of Lake Erie (Figure 1). The refuge encompasses a large complex of Great Lakes coastal wetlands that have been extensively altered by human activity. As such, it contains many wetland pools that are isolated from the neighboring riverine and lacustrine systems by earthen dikes. Although these wetland, creek, and lake environments are spatially close, they differ in their chemical and physical characteristics. Consequently, they support similar, but not identical, communities of large-bodied fish. Common Carp (Cyprinus carpio) and Bowfin (Amia calva) are common throughout, while Northern Pike (Esox lucius) and Yellow Bullhead (Ameiurus natalis) predominate in wetlands, and Gizzard Shad (Dorosoma cepedianum) and Smallmouth Buffalo (Ictiobus bubalus) are more common in the creek environment (Kowalski et al., 2014). These species differ in size, physiology, habitat preferences, and life history dynamics, contributing to a complex landscape-scale fish assemblage that requires a diversity of habitat types over time (Jude and Pappas, 1992).

In March 2011, a water-control structure with two 1.2 m diameter culverts was opened between a 40 ha diked wetland pool and the drowned mouth of Crane Creek, restoring hydrological connection to Lake Erie for the first time in over 50 years (Figure 1). This structure allows short and long-term water-level changes in Lake Erie (e.g., seiches, annual fluctuations) to propagate into the wetland. It also enables new patterns of fish emigration and immigration between the wetland, creek, and lake, expanding the spatial resources available for foraging, reproduction, predator avoidance, and milling behaviors. While the water-control structure remained open for the duration of the project discussed here, carp grates with a 7 cm horizontal spacing between vertical bars were installed for a variable number of weeks during carp spawning season each spring (Table 1). These grates are designed to limit Common Carp immigration into the wetland pool but may also affect large non-target species unable to pass through the grates.

Sonar Deployment

Acoustic observations were made following methods detailed in Eggleston et al. (2020). Briefly: a Dual-frequency Identification Sonar unit (DIDSON 300, Sound Metrics: Bellevue, Washington, U.S.A.) was deployed 56 times for approximately 48 hours between March 2011 and November 2014 (Table 2). These observations were made intermittently, approximately every two weeks, depending on factors such as ice cover, inclement weather, equipment readiness, and operational status of the United States government.

At each deployment, the DIDSON unit was mounted to a permanent metal truss and submerged approximately 60 cm below the surface of the wetland pool. It was positioned approximately 2 m southeast of the water-control structure and oriented so that its beams intersected water flow between the creek and wetland in a perpendicular manner (see Figure 1 and in-text description in Eggleston et al., 2020). From this position, the full width of the channel leading from the wetland into the control structure could be ensonified, so that objects present in the eastern portion of the channel appeared in the foreground of echograms while those in the western portion appeared in the background. This arrangement allowed us to capture nearly all submerged objects moving in or out of the wetland pool. Precision adjustments were made using mounting hardware at each deployment to minimize variability in the field of view between observations. Digital images were collected at a rate of 10 frames s^-1 and cached to an external hard drive at one-hour intervals in “*.ddf” format. All observations were made in high-frequency identification mode (1.8 MHz) to optimize image clarity. A typical echogram is presented in Figure 2.

Digital Image Processing

Sonar data were processed according to methods previously described in Eggleston et al. (2020). To ensure comparability between observational units, sonar echogram files shorter than one hour (resulting from equipment malfunction or early deployment termination) were excluded from subsequent analyses. Additionally, digital echogram files were manually reviewed at 10X speed, and any files containing a continuous blackout of more than six minutes, a shift in the field of view (resulting from unusual water flow or wildlife interference), human activity, or reduced image clarity were eliminated from the analytical set.

A combination of two software applications, a commercial program for processing hydroacoustic data (Echoview v6.0, Echoview Software Pty Ltd: Hobart, Tasmania, Australia) and a purpose-built application from Milne Technologies (Keene, Ontario, Canada), was utilized to derive estimated fish counts from each echogram file. The Milne Technologies application automated sequential processing of multiple echogram files, significantly reducing the operator time required to process thousands of hours of echogram data. Prior to use, Echoview was calibrated with our dataset by comparing computerized fish counts with those obtained manually from echograms of a similar origin. Specifically, adjustments were made to default parameters for major and minor axes, exclusion distance, missed ping expansion, and track acceptance to optimize agreement between Echoview and manual counts. Echoview was used to identify, count, and characterize the direction of movement (towards or away from the connecting structure) for all fish greater than 8 cm in length in each hour-long echogram. While the listed adjustments minimize error and generated values are generally regarded as useful, these values are imperfect and can vary with respect to target density, species, and path geometry (Hightower et al., 2013; Hughes and Hightower, 2015). Moreover, the figures presented here reflect detections made within the ensonified area and may not be representative of the wetland pool overall. As a result, the abundance values presented herein can be regarded as estimates that are proxies for fish populations and movement but not precise counts of individual fish.

Statistical Analysis

All statistical analyses were performed in the R statistical environment (v4.4.0; R Core Team, 2024). Daily mean flood stage values and hourly means for all water quality parameters (pH, conductivity, water temperature, dissolved oxygen, and turbidity) were obtained through ServCat, the USFWS’s online data repository. These data were calculated from measurements from a gage located near the wetland pool within Crane Creek (41°37'21.347" N, 83°12'40.758" W). All environmental data were rated at least “fair” by the USFWS, except for water quality information from 2013 that was rated as “poor.” Despite this, the 2013 data constituted the best available and most comparable information.

We transformed flood stage values by centering around the global mean and scaling by the observations’ standard deviation. We described hourly temporal variation using astronomical time since it has been repeatedly shown to play a major role in shaping fish behavior in the Great Lakes and other temperate North American freshwater systems (Helfman, 1981; Forsythe et al., 2012; Gorman et al., 2012; Kough et al., 2018). All figures depict fish activity using the local time at observation. However, to describe diel variability in model analysis, we also calculated the number of hours since an observation’s preceding sunrise as an integer between 0 and 23 using the R package suncalc (Thieurmel and Elmarhraoui, 2022). For instance, when considering the 24 observational periods that follow a 07:15 sunrise, the hour-long observation beginning at 10:00 was assigned a value of 3 (hours following sunrise) and an observation beginning at 03:00 assigned a value of 20. To describe seasonal variation, we used a similar approach to calculate the number of weeks that had passed since an observation’s preceding vernal equinox.

We calculated means and 95% confidence intervals describing fish detections per hour of acoustic surveillance within each hour since the preceding sunrise, week since the spring equinox, deployment event, year, and heading direction. As fish use of this wetland pool is likely the result of multiple, interactive ecological and environmental factors, we implemented generalized linear mixed model (GLMM) analysis using the R package glmmTMB v1.1.9 (Brooks et al., 2017). Specifically, we implemented a type II negative binomial mixed model (V = µ + µ²/ϕ) to assess the influence of temporal and environmental variables on the likelihood of fish detection. We selected this model type because the fish count data were not normally distributed (Shapiro-Wilks, W= 0.837, P < 0.001) and exhibited high dispersion (x̄ = 1 700, s² = 2 790 706). To account for the serial correlation inherent in the data structure, we included a lag 1 autoregressive covariance structure, grouped by deployment event, in all constructed models.

We first tested for differences among years and between rates of immigration and emigration using all available data (N= 2 022) and single fixed effects. Additionally, detection frequency data could be matched with environmental data for 1 119 hours of observation. Since temporal influences on fish activity at diel and seasonal scales are well documented in similar systems (e.g., Emery, 1973; Helfman, 1981) and visually apparent upon inspection of our data, our simplest model described fish detection frequencies using fixed effects for weeks since the vernal equinox, hours since the last sunset, and the interaction of these terms. We constructed additional models using single term additions for each environmental variable to the temporal-effect-only model (Model 1) and used Akaike Information Criteria (ΔAIC, Akaike, 1974) to identify those terms that improved model fit. We then created additional models parameterized with all combinations of these fit-improving environmental variables as main effects in addition to temporal terms. Accordingly, we occasionally included both water temperature and dissolved oxygen in same model despite their physical relationship, as each parameter can affect fish behavior independently (Zhang et al., 2024). We disregarded models that failed to converge. For subsequent analyses, we considered our best temporal model and any more parsimonious models that included environmental variables. We calculated Nakagawa’s marginal R²_GLMM (Nakagawa and Schielzeth, 2012) as a partial description of model fit using the R package MuMIn (Bartoń, 2024), and we considered parameters whose coefficient estimates did not overlap zero in at least one of these best-fit models to be potential determinants of fish activity.

Sonar deployment and fish detections

We successfully characterized fish use of the wetland for 2 022 hours of sonar data spanning 56 individual deployments (Table 2). Daytime hours were slightly overrepresented (N = 1 002) compared to night (N = 529) and crepuscular periods (N = 491). Between 2 and 68 hours of echograms were successfully processed per deployment (mean = 37 ± 14). From these observations, we obtained occurrence and heading information for 3 436 118 total fish (Table 3). In each hour-long echogram, we detected between 3 and 9 353 fish.

Direction of movement

We consistently detected more fish entering the wetland pool (moving from right to left across the echograms) than exiting it towards Crane Creek and the broader Lake Erie basin, However, these differences were modest, comprising only 1% - 5% of total detections (Tables 3 and 4). When tested using a GLMM, differences in mean detection by direction of travel were found to be small but statistically significant (Table 5). Nonetheless, model marginal pseudo-R² was extremely low (Table 5), indicating that patterns of variation in fish activity directed towards the wetland pool were closely linked to those of fish moving towards open water (Figures 3 and 4).

Temporal Variables

Environmental variables

Three models including environmental variables were identified as more parsimonious than the temporal-effects-only model (Table 8). These models incorporated flood stage, water temperature, and/or dissolved oxygen as fixed effects alongside the temporal terms. Both water temperature and dissolved oxygen yielded non-zero coefficient estimates at a standard significance level (P < 0.05), while flood stage produced a non-zero coefficient at a less conservative significance level (P < 0.10). Both water temperature and flood stage had positive relationships with fish detections, whereas dissolved oxygen was associated with a reduction (Table 7). Other environmental variables tested (pH, conductivity, and turbidity) were not identified as meaningful sources of variation by this analysis.

Daily temporal variation

Changes over time in the compositions of Great Lake wetland fish communities are well documented (Kowalski et al., 2014; Langer et al., 2016; Reisinger et al., 2019; Diller et al., 2022). These changes arise from temporal differences in species-specific life histories governing use of wetlands. As new individuals or populations enter or exit the wetland, we expect to observe corresponding increases or declines in fish detections, enabling us to infer changes in wetland use over time. The resulting description of temporal fluctuations in fish activity within coastal wetlands of the Great Lakes has potential value in the monitoring and management of these resources.

Overall, this area of the reconnected wetland was used continuously by fish while the fish-passage structure was open. Although fish use of restored Lake Erie coastal sites has been previously documented (Steinhart et al., 2017; Stott and Miner, 2022), this project provides evidence of rapid pool use beginning immediately after initial reconnection (Figure 4). Moreover, use of this area exhibits pronounced diel rhythms, with greatest activity observed during the early morning hours and lowest activity overnight (Table 7, Figures 3 and 5). While comparisons of night and daytime freshwater fish communities in temperate regions of eastern North America and the Great Lakes suggest that some form of diel variability is the norm, exact patterns of high and low usage appear to be system specific. For instance, previous studies of vegetated coastal zones show that measures of community diversity often fluctuate throughout the day, but the direction of these shifts is variable. Some studies, using seine nets or electrofishing, report greatest richness or diversity at night (Pierce et al., 2001; McKenna, Jr., 2008; Midwood et al., 2016; Ross et al., 2016), while others, using manual surveys or trap nets, document the opposite trend (Keast et al., 1978; Shoup et al., 2004). Nighttime communities are also often more abundant (Pierce et al., 2001; Midwood et al., 2016; Ross et al., 2016), but these relationships can be mixed—even when evaluated with similar gear types (McKenna Jr., 2008). In addition, variation in these trends within study systems has been linked to factors such as community composition, habitat type, or time of year (Keast et al., 1978; Larocque et al., 2020).

In contrast to most of these findings, we typically detected the lowest fish activity and inferred lowest abundances during nighttime hours (Figure 3 and Table 7). Notably, elevated numbers were usually observed only for a few hours surrounding and following sunrise, with detection rates considerably reduced from late morning through the evening. This short-term variation could obscure some diel patterns in fish use, particularly if the entire diurnal period is characterized by a sample taken later in the day. Such a method could overlook an earlier period of high use, contributing to some of the inconsistencies between our finding and previous studies. This possibility reinforces the need for high temporal coverage when studying diel variation in fish activity.

Although the full complexity and breadth of ecosystem dynamics producing these diel rhythms are best explored using other means (e.g., Gorman et al., 2012; Marcy-Quay and Kraft, 2021), we found modest evidence that diel changes in water oxygenation could contribute to these patterns. Our analysis suggested a negative relationship between dissolved oxygen concentration and the rate of fish detections (Table 7). Although complicated by intermittent seiche-driven water exchange, this relationship could arise from diel oxygenation cycles typical of highly vegetated, warm bodies of shallow water common in coastal wetland systems. These cycles are characterized by sizable increases in dissolved oxygen accumulating as a byproduct of photosynthesis between sunrise and sunset. In contrast, many of the fish species likely to inhabit this wetland are either moderately tolerant of or specialized for low-oxygen environments and tend to be less common in well-oxygenated areas (Ludsin et al., 2001; Arend et al., 2010; Kowalski et al., 2014). Consequently, these fish may respond to afternoon and early evening increases in photosynthetically-derived oxygen by retreating towards deeper, unvegetated areas beyond the ensonified zone or exiting this heavily-vegetated pool altogether, leading to fewer fish detected later in the day. A few highly abundant taxa of this type could drive the observed pattern, as is the case in other systems where just a few abundant taxa play a major role in driving diel trends (Keast et al., 1978, Pierce et al., 2001). Such a coordinated entry by species that are absent or less common in other, previously-studied systems could also explain inconsistencies between those studies and the diel patterns we document here. However, more taxonomically-sensitive methods would be necessary to confirm this hypothesis or identify the taxa involved. Moreover, given this limitation concerning species identity, we cannot know whether these detections and related trends represent diverse or homogenous communities in terms of species, functional groups, or invasive status, nor how these patterns might change over time.

Despite this, the information presented here could be valuable in designing and planning management or monitoring programs in order to minimize off-target impacts and maximize coverage of species of interest. Our data suggest that short-term management actions occurring over minutes or hours could be optimized through the understanding of diel cycles in fish activity. Incorporating this information could help mitigate the negative impacts of disruptive management actions, like chemical applications or operation of the water-control structure, by scheduling these activities during late afternoon or early evening when fish activity is lower. Alternatively, management activities that benefit from greater representation of the fish community, such as monitoring or stock assessments, could be strategically targeted for early morning hours when fish are more active. Similarly, the timing of data collection events should also consider these diel patterns to maximize comparability among observations and most effectively address project goals. Although annual variability in these diel patterns is an apparent feature of this dataset, certain trends (e.g., high early-morning activity) are repeated and represent a useful starting point for planning purposes, which could be paired with more site-specific knowledge or observations, as appropriate. Additionally, this information could be combined with taxonomic data to target management operations further. However, it is important to note that there were no significant time periods when the fish-passage structure was entirely devoid of fish, and thus a purely temporal mitigation strategy is unlikely to eliminate undesirable impacts completely. Similarly, our analysis of fish movement in the structure does not represent the total number and diversity of fishes in the wetland itself.

Long-term temporal variation

Seasonal changes in the abundance or diversity of fish communities in Great Lakes coastal wetlands are not unusual (Keast et al., 1978; Chubb and Liston, 1986; Bhagat and Ruetz, 2011, Diller et al., 2022). This study provides evidence of pronounced seasonal changes in the estimated abundances of large-bodied fishes utilizing a reconnected wetland pool (Table 7, Figure 4). We measured sizeable and annually recurrent increases in fish detections during late spring. These elevated detection rates usually persisted for 4-6 weeks before returning to early spring levels in midsummer. These patterns may be underlaid by community change driven by changes in biotic and abiotic features of the wetland environment.

High springtime detections, indicating significant immigration, may be partially influenced by changes in abiotic conditions. Substantial seasonal variations in physical and chemical parameters occur in this reconnected wetland, affecting its suitability for specific fish species (Kowalski et al., 2014; Carter et al., 2022). In particular, we found modest evidence connecting rising water level and temperatures to higher rates of fish detection (Tables 7 and 8). Both factors have been found to play a role in structuring use or community composition in other Great Lakes wetlands (Wismer and Christie, 1987; Fracz and Chow-Fraser, 2012; Langer et al., 2018; Stott and Miner, 2022). Water levels are usually highest in spring, coinciding with springtime warming and high rates of fish activity (Figure 4). Increased water levels enhance flow through the connecting structure and warming temperatures facilitate the elimination of ice within the channel connecting wetland and creek. However, although the resulting wetland accessibility usually begins in March, peak fish activity was not observed until late April or May (Figure 4). This suggests springtime influx is not solely a result of habitat availability but may also be a response to other stimuli. Spawning conditions are likely key triggers of immigration, as several large-bodied wetland spawners are present in this system, including Northern Pike, Bowfin, Common Carp, and Brown Bullhead (Ameiurus nebulosus) (Hartley and Herdendorf, 1977). Another possible trigger involves small cyprinids that can be highly abundant in Great Lakes coastal wetlands during the early spring (Diller et al., 2022). Given their size, these forage fish would not be detected individually by our method. However, they could attract large-bodied piscivores as they accumulate in high densities later in the spring. Ultimately, this study design cannot adequately identify these factors, and further research is needed to identify the specific triggers that influence the timing of springtime influx.

Undoubtedly, Common Carp constitute a portion of spring immigration. This abundant, non-native, and frequently nuisance species can create dramatic and undesirable effects in wetlands (Lougheed et al., 1998; Kowal et al. 2022; Moore and Mandrak, 2024). They often make up a major and conspicuous portion of springtime influx into Great Lakes coastal wetlands as they use these areas for spawning (Bajer and Sorensen, 2009; Banet, et al., 2021; Watkinson et al., 2021; Piczak et al., 2023a). However, carp grates, designed to prevent carp of spawning size from accessing the restored wetland without affecting other species, were installed annually across the channel linking the restored wetland habitat and Crane Creek (French et al., 1999). This installation should have considerably reduced the number of large Common Carp able to enter the wetland but may have had small impacts on other species of large fish. As a result, it is possible that similar, ungated coastal wetlands could experience even greater spring influxes of large-bodied fish (including Carp) than are described here.

We detected the greatest number of fish in this wetland during late spring and early summer, however, there are no deployment periods in which large-bodied fish were entirely absent (Figure 4). Deployments from mid-summer through late fall exhibit relatively stable detection rates, indicating reduced change in the population of large-bodied fishes over this period compared to late spring (Figure 4). In general, Great Lakes coastal wetlands with large anthropogenic impacts are inhabited by distinctive fish species, possessing functional traits well-suited for degraded wetland habitats (Trebitz et al., 2009; Johnson et al., 2025). The stable pattern of detections in early spring and from late summer onward may correspond to a year-round assemblage of these resident species. This guild of fish is better able to exploit wetland habitat characterized by large seasonal changes in water quality (Bhagat et al., 2007; Arend et al., 2010; Parker et al., 2012). An assemblage of this type, highly tolerant of low oxygen, would also contribute to the observed negative association between dissolved oxygen concentration and fish activity (Piczak et al., 2023b).

Seasonal variability in the use of this restored wetland has significant implications for management, especially for multi-day or multi-week projects such as hydrological manipulation or the exclusion of undesirable species. These actions could be strategically scheduled outside of the late spring fish influx to minimize impacts on fish populations or maximize exclusion of undesirable species (e.g., Common Carp). Conversely, if the goal is to manage or monitor all fish utilizing the wetland, executing these management actions during late spring could capitalize on the concentrated presence of fish in this area even though it is not clear how the management activities themselves may influence activity patterns. Monitoring activities could also target or avoid these seasonal cycles to ensure comparability between observations and comprehensive coverage of the wetland’s fish community. Additionally, if hypothesized species turnover is indeed occurring, this temporal information could be combined with additional data to target specific fish species or functional groups (e.g., Stott and Miner, 2022; Piczak et al., 2023b).

Finally, we found no evidence of sustained directional trends in annual fish activity during this period. The above seasonal and diel patterns structuring fish remained largely stable over these four years (Table 6 and Figure 4). This stability is consistent with findings from coastal wetlands along Lake Michigan’s eastern shoreline where no major year-to-year variation was observed (Bhagat and Ruetz III, 2011). However, we do observe specific instances of unique fish use dynamics. For example, we documented a sudden, large, and protracted reduction in fish activity that lasted throughout the entire deployment event between May 17-18, 2011. This decline was not observed in subsequent years and may have been linked to remarkably rough weather, strong currents, high water, and elevated turbidity levels during that time. While these conditions did influence fish activity, they do not appear to be recurrent in nature. Although not a major feature of these findings, the importance of weather-related perturbations in structuring chemical and biological dynamics of this system has been previously noted (Baustian et al., 2018), and significant weather events are likely to become more frequent in the context of a changing climate (Zhang et al., 2020; Hicke et al., 2022). Moreover, while we did not find evidence of multi-year directional trends during the four-year interval presented here, this does not preclude the possibility of decadal or century-scale trends in fish activity (e.g., Ludsin et al., 2001).

Limitations

While this project does characterize some general patterns of wetland pool use, several fish behaviors or methodological considerations make it likely that we overestimate fish abundances through repeated detections or misattribution (Holmes et al., 2006). Repeated counts of individual fish, generated by milling behavior, returns to the ensonified area, or broken fish tracks generated by the analysis software likely contribute to this pattern. In addition, improper detections, generated by acoustic noise, waterborne debris, or high turbidity also reduce count accuracy. While both these factors would tend to affect immigrating and emigrating counts equally, we also regularly detected slightly more fish entering the wetland than exiting it (Tables 3 and 4). Such a pattern could arise from subtle, but pervasive, biases arising from flow regimes. For instance, immigration numbers may be slightly inflated if incoming water velocity is high and consistently pushes debris into the wetland that is misclassified as fish detections. Less debris in the water column as the wetland pool drains to Crane Creek could be generating fewer sonar detections. Regardless, manual counts of DIDSON imagery have been shown to produce comparable counts to the method implemented here (Eggleston et al., 2020). Ultimately, while our data are most likely imperfect estimates of actual fish use, the repeated nature of these fluctuations suggest that they represent real biological phenomena.

We assigned each fish a direction of movement, either towards the wetland pool or into Crane Creek, while analyzing the sonar data. We anticipated that this would allow us to describe temporal trends in immigration and emigration separately, as performed in previous work (Stott and Miner, 2022). However, the performance of acoustic monitoring varies based on analytical method and deployment location or configuration (Boswell et al., 2008). In this case, we did not find strong evidence attesting that our differentiation was successful. Most tellingly, mean fish movement towards and away from the connective structure were very highly aligned at multiple timescales (Figures 3 and 4). If this method had successfully distinguished between emigrating and immigrating fish, we would expect to observe distinct abundance patterns between immigrants and emigrants corresponding to coordinated, but temporally segregated, movement in and out of the wetland. While it is possible that any effect of temporal specialization was masked by considering the community in aggregate or at this temporal resolution, other studies have documented variations in habitat use on similar timescales (Funnell et al., 2023; Peterson and Grubbs, 2023). In the future, a slightly different computational approach (e.g., only assigning directional movement to fish that travel a minimum distance in a single direction) may be more successful in describing directional movement.