Introduction

In polygamous mating systems, behaviors such as signaling and displaying by males are performed to portray fitness and social dominance, attract females, and secure mating opportunities (Emlen and Oring 1977; Anderson and Simmons 2006). Mate attraction is directly linked to the preference and likelihood of successfully reproducing, therefore individuals should adopt signaling strategies that maximize effective and safe communication in a dynamic environment (Lima 1998; Oseen 2002). Studies across a wide diversity of species indicate that animals alter signaling behaviors based on environmental factors, presumably to facilitate improved communication (Uy and Endler 2004; Leal and Fleishman 2004; Wilgers and Hebets 2011; Brenner et al. 2019). The acoustic adaptation hypothesis posits that animals communicating vocally should adapt their vocalizations to local conditions to optimize signal transmission (Ey and Fischer 2009; Gonzalez et al. 2018). A key component of the acoustic adaptation hypothesis is that vegetation structure affects sound transmission differently (Morton 1975; Ey et al. 2009), so certain land cover types are more optimal for signal transmission, and animals should vocalize more in these areas to increase fitness (Pfaff et al. 2007; Seehausen et al. 2008).

Predation risk has potential to influence behavior of prey, and the risk reward theory predicts that species who signal must do so in a way that benefits obtained from signaling outweigh predation risk and potential death (Jennions et al. 1997; Zuk and Kolluru 1998). Furthermore, the landscape of fear theory infers that prey will adopt anti-predator behaviors in areas on the landscape associated with increased risk (Laundre et al. 2001; Moll et al. 2017; Creel 2008). For example, prey species will adjust the frequency of signaling and distribute themselves spatially and temporally on the landscape in a way to avoid perceived risk (Burk 1982; Hedrick 2000; Lohrey et al. 2009). Collectively, these hypotheses offer possible explanations for observed variation in animal signaling and vocalizations as they look for optimal land cover for sound transmission while minimizing risk of predation by altering frequency of vocalizations in high risk situations or moving to areas of reduced risk to call.

The wild turkey (Meleagris gallopavo) is a non-migratory upland game bird indigenous to North America whose mating strategy is a form of polygamy, whereby females select males who are competing for mating opportunities via signaling behaviors (Healy 1992; Krakauer 2008). Males use vocalizations (hereafter gobbles) and visual cues (e.g., strutting) to attract females, establish and maintain dominance hierarchies, and compete with male conspecifics for breeding opportunities (Bevill 1973; Healy 1992). Male wild turkeys are primarily hunted during the spring reproductive season; spring harvest is the primary cause of mortality for males (Hughes et al. 2005; Chamberlain et al. 2012), and hunting activity can influence male movements and signaling behaviors (Gross et al. 2015; Collier et al. 2017; Chamberlain et al. 2018; Wightman et al. 2019; Wakefield et al. 2020a). Furthermore, a suite of natural avian and mammalian predators prey on turkeys (Miller and Leopold 1992; Hughes et al. 2005), and growing evidence suggests that many of these predator populations are currently increasing in abundance (Prugh 2009; Sauer et al. 2013). For example, coyotes (Canis latrans) have dramatically increased their geographic range during the past few decades, and now are ubiquitous throughout North America (Hody and Kays 2018). Although coyotes are not believed to remove significant numbers of adult wild turkeys (Moore et al. 2008; Ward et al. 2018), they do depredate turkeys and the potential that coyote presence on the landscape may cause various behavioral changes (e.g., vigilance, habitat shifts) in turkeys is plausible given growing literature noting indirect effects causing behavioral changes in similar predator prey systems(Bolnick and Preisser 2005; Creel et al. 2008; Laundre et al. 2010; Sheriff et al. 2020). Aside from land cover types associated with reduced sound attenuation and predation risk, both male and female wild turkeys alter patterns of habitat selection as the spring reproductive season progresses (Miller and Conner 2007; Martin et al. 2012; Little et al. 2016a), which could influence variation in gobbling activity across portions of the landscape.

Hearing turkeys gobble increases quality of hunting and hunter satisfaction (Little et al. 2000; Casalena et al. 2011; Isabelle and Reitz 2015), Temporally, gobbling activity is most influenced by nesting phenology/receptivity of females in the absence of hunting (Chamberlain et al. 2018; Wakefield et al. 2020a).However, in the presence of hunting, season start date and the removal of males are most influential (Chamberlain et al. 2018; Wightman et al. 2019; Wakefield et al. 2020a). On areas with an open hunting season, gobbling activity is reduced as the hunting season progresses, resulting in reduced gobbling compared to non-hunted sites (Wightman et al. 2019; Wakefield et al. 2020a). However, previous studies on hunted sites noted substantive spatial and temporal variability in gobbling activity (Colbert 2013; Wightman et al. 2019; Wakefield et al. 2020a), and Wightman et al. (2019) suggested that this variation is greater compared to non-hunted sites. Collectively, these findings suggesting that certain locations on the landscape have characteristics that facilitate greater gobbling activity. However, to our knowledge, no study has attempted to determine what landscape characteristics and other environmental factors may predict spatial variation in gobbling activity. Therefore, our objective was to investigate potential influences of predation risk, turkey use, and land cover type on spatial and temporal trends in gobbling activity. We predicted that areas of decreased predation risk, increased predicted probability of turkey use, and open land cover types would all positively influence gobbling activity but at varying degrees. Specifically, given the landscape of fear model and the risk allocation hypothesis, we predicted that risk from hunters would have the greatest impact on gobbling activity and would be greater compared to coyotes, as hunters are an ephemeral predator on the landscape for a relatively short period of time and represent high risk and increased lethality compared to coyotes.

Material and methods

Study site

During 2018–2019, we collected data on 2 Wildlife Management Areas (WMAs) approximately 8 km apart in the Piedmont region of Georgia, USA (Fig. 1). The first was Cedar Creek WMA, a 16,187-ha area located in Jasper, Jones, and Putnam counties owned by the United States Forest Service (USFS) and managed in partnership with the Georgia Department of Natural Resources-Wildlife Resources Division (GADNR). Cedar Creek WMA (CCWMA) land cover types consisted of approximately 40% (6,475 ha) upland planted pine stands of mostly loblolly pine (Pinus taeda), 49% (7,932 ha) bottomland oak hardwoods containing white oak (Quercus alba), sweet gum (Liquidambar styraciflua), yellow poplar (Liriodendron tulipifera), hickory (Carya spp.), and other oak species (Quercus spp.), 2% (323 ha) of interspersed mixed pine-hardwood forests, < 1% open water, 3% (486 ha) roads, 3% (486 ha) open areas planted/maintained for wildlife, and 4% (647 ha) shrub/scrub consisting mainly of sweet gum and briars (Rubus spp.). In 2018, a turkey hunting season was open to the public from 24 March to 15 May, whereas in 2019 it spanned from 6 April–15 May.

Fig. 1
figure 1

Location and land cover features of B.F. Grant and Cedar Creek Wildlife Management Areas, and surrounding property in Georgia, USA, 2018–2019

We also collected data on the 4,613-ha B. F. Grant WMA (BFG) located in Putnam County, Georgia. The BFG was owned by the Warnell School of Forestry and Natural Resources at the University of Georgia and managed in partnership with the GADNR. Land cover types consisted of approximately 30% (1,384 ha) planted loblolly pine forests intensively managed for timber, 44% (2,230 ha) hardwood forests similar to CCWMA, 1% (46 ha) mixed pine-hardwood forests, < 1% open water, 3% (138 ha) roads, 18% (830 ha) open areas consisting primarily of mixed fescue (Festuca spp.) agricultural fields for cattle grazing and rye grass (Lolium spp.) fields for hay production, and 4% (184 ha) shrub/scrub consisting mainly of sweet gum and briars. Turkey hunting season on BFG was split into 3 parts, the first a youth hunters only hunt which occurred from 24 March–1 April in 2018 and 23 March–31 March in 2019. The second hunt was an 80-person quota from 2 April–8 April in 2018 and 1 April–7 April in 2019. The final hunt was open to the general public and occurred 9 April–15 May in 2018 and 8 April–15 May in 2019. Climate on our study sites was characterized by hot, dry summers and cool, wet winters. Elevation ranged from 80 to 250 m and the average elevation across our study areas was 156 m.

Gobbling data collection

To measure gobbling activity, we deployed autonomous recording units (ARUs; Song meter model SM4 + ; Wildlife Acoustics, Concord, MA, USA) from 1 March–30 June during 2018–2019. We scheduled ARUs to record all ambient sound from 0500 h until 1100 h (hereafter daily gobbling), which encompassed > 85% of daily gobbling activity (Wightman et al. 2019). We separated ARUs by > 300 m in areas observed to have turkey activity (Colbert et al. 2015; Wightman et al. 2019). We caged ARUs and placed them 3 m off the ground in trees to reduce human interference and placed a microphone 9 m off the ground to maximize calling detection range (Colbert et al 2015). We used a Convolutional Neural Network (CNN) to autonomously search for turkey gobbles and identify potential occurrences of gobbles at each ARU (Wightman et al. 2022). We implemented the CNN in Python with the Keras library (Chollet 2017) using a backend of the open-source TensorFlow software developed by Google. For each potential gobble identified by the CNN, a record was created containing call location in the spectrogram, date, time stamp, and a 3 s sound file of the potential gobble. We auditorily verified all identifications and classified each as a true or false gobble, producing daily counts of gobbles at each ARU.

Hunter risk variables

Spring harvest is the primary cause of mortality for male wild turkeys, and hunter activity is known to influence various aspects of wild turkey behavior (Vangilder 1992; Chamberlain et al. 2012; Gross et al. 2015; Wightman et al. 2019). Contemporary studies have identified areas of increased hunter use on the landscape, which are highly correlated with distance to the nearest road (Diefenbach et al. 2005; Karns et al. 2012; Gerrits et al. 2020). Therefore, to investigate potential effects of hunter activity on gobbling activity at each ARU, we calculated the nearest distance to primary and secondary roads using the Euclidean distance tool in ArcGIS 10.5.1 (Environmental System Research Institute, Inc., Redlands, CA, USA). We obtained road data within both WMAs from GADNR and used 2019 USGS Tiger/Line data (Topologically Integrated Geographic Encoding and Referencing) for roads outside the WMAs. We categorized roads as primary if they were paved/graveled and vehicle access was not limited, whereas secondary roads were unpaved gravel and/or logging roads where vehicle use was prohibited (Gerrits et al. 2020). Further, some areas of our study sites were inaccessible from a primary road due to private property between the road and ARU, therefore we also calculated the nearest distance of public access where the general public could park a vehicle and access public property and each ARU, assuming the farther the ARU was from public access, the less hunting activity would be associated with that area (Gerrits et al. 2020). Likewise, we assumed that private lands would represent areas of reduced hunter activity due to decreased hunter density compared to public land (Root et al. 1988; Small et al. 1991; Haus et al. 2019), so we calculated distance to private property from each ARU.

Coyote risk variable

During 2018–2019, 36 coyotes were captured by a professional trapper using MB450 foothold traps during January–March (Chamberlain et al. 2021). Before release, coyotes were fitted with GPS collars with Iridium capabilities (Advanced Telemetry Systems G5-2A GPS-Iridium, Lotek LiteTrack Iridium 360). We programmed collars to record 1 location every hour, but we subset the data to match that of turkeys (0500–2000, 1 March–30 June, see below) as we were only interested in times when coyotes and turkeys could be directly interacting (excluded times at night when turkeys roost in trees; Porter 1978; Byrne et al. 2015). Coyote capture, handling, and marking procedures were approved by the Institutional Animal Care and Use Committee at the University of Georgia (Protocol number A2014 08-025-R2).

We used spatial locations of coyotes to construct a resource selection function (RSF; Manly et al. 2002; Online Resource 1: resource selection analyses) for coyotes, based on a suite of landscape covariates (Online Resource 1: landscape covariates). We calculated the above RSF and parameter estimates (Online Resource 2: Table 1) to develop a relative probability of use for coyotes for each 30 m × 30 m pixel on our study site before, during, and after the anthropogenic disturbance of hunting. To determine the average predicted probability of use at each individual ARU, we created a 250 m buffer using ArcMap 10.3.1 (ESRI, Redlands, CA, USA). We buffered at 250 m because that distance approximates the known distance at which ARUs can effectively record gobbling activity (Colbert et al. 2015; Wakefield et al. 2020b). We then calculated the average predicted probability of use at each ARU by averaging the value for every 30 m × 30 m pixel within or overlapping the 250 m buffer. The average predicted probability of coyote use at each ARU was our proxy for coyote risk in subsequent modeling.

Table 1 List of Bayesian regression models and parameters used to estimate effects on gobbling activity of Eastern wild turkey on Cedar Creek and B.F. Grant wildlife management areas in Georgia, USA, during March–June 2018–2019. Each model includes an interactive term for stage (before, during, and after hunting) and a random intercept term to account for variation across autonomous recording units. Models were compared by marginal and conditional R2 values and ranked using expected log pointwise predictive density (ELPD) and leave-one-out-information-criterion (LOOIC)

Turkey use variables

During 2018–2019, 111 (81 F, 30 M) turkey were captured during January–February using rocket nets baited with corn. Upon capture, birds were sexed, aged, and fitted with a remotely downloadable GPS-VHF radio transmitter (Guthrie et al. 2011; Biotrack Ltd., Wareham, Dorset, UK). Transmitters were programmed to record 1 location nightly (23:58:58) and daily (0500–2000) every hour, until the battery died or the unit was recovered (Byrne et al. 2014). We located turkeys > 3 times per week using handheld antennas and VHF receivers to monitor survival while we downloaded locations from each turkey at least once per month. If an individual was found dead, GPS data were downloaded upon recovery and truncated to only include points taken while the individual was alive based on our VHF tracking. We used GPS data collected from 1 March–30 June for subsequent analyses. Turkey capture, handling, and marking procedures were approved by the Institutional Animal Care and Use Committee at the University of Georgia (Protocol number A2016 04-001-Y2-A0 and A2019 01-025-R2).

Both male and female turkeys alter resource selection as the spring reproductive season progresses (Miller and Conner 2007; Martin et al. 2012; Little et al. 2016a), hence areas on the landscape with increased male and female use could be associated with areas of increased gobbling activity. Therefore, we used spatial locations of turkeys to construct a resource selection function (RSF; Manly et al. 2002; Online Resource 1: resource selection analyses) for male and female turkeys, based on a suite of landscape covariates (Online Resource 1: landscape covariates). We calculated the above turkey RSFs and parameter estimates (Online Resource 2: Table 2) to develop a relative probability of use for male and female turkeys at each 30 m × 30 m pixel on our study site before, during, and after the anthropogenic disturbance of hunting. We used the same methodology above for coyote risk to calculate the average predicted probability of use for both males and females at each ARU before, during, and after hunting.

Land cover variables

We generated individual raster layers for 6 vegetation types thought to influence turkey behavior (Yeldell et al. 2017; Chamberlain et al. 2020). Land cover types included pine, hardwood, shrub, open, mixed, and water/wetland. We quantified land cover types using 30 m × 30 m raster layers from the United States Department of Agriculture (USDA), National Agricultural Statistical Service for 2018 and 2019 (USDA 2018–2019). The USDA cropland data layers provided annual geospatial updates to land cover data, so we were able to account for changes in forest cover, specifically clear-cuts resulting in open and shrub areas due to active timber harvesting on both sites. We used the Euclidean distance tool in ArcGIS 10.5.1 (Environmental System Research Institute, Inc., Redlands, CA, USA) to calculate the minimum distance from each ARU to each land cover type. We then used these distance metrics for subsequent analysis instead of a classification or categorical approach (Conner et al. 2003).

Data analysis

To evaluate the influence of predation risk, turkey use, and landscape characteristics on gobbling activity, we subset our gobbling data into counts of gobbles at each ARU during 3 stages (before, during, and after hunting season) to reflect potential effects of hunting on gobbling activity. We then divided the total number of gobbles detected during each stage by the number of days recorded by each ARU. We subsequently rounded to the nearest gobble to get an average daily gobbling count (hereafter gobbling activity) at each ARU during each stage, which allowed us to account for uneven sampling days recorded by ARUs where data were missing due to ARU malfunction. Previous research noted that hunting and removal of males lead to decreased gobbling activity during the hunting season, and substantive variability in gobbling activity across ARUs (Online Resource 2, Fig. 1; Wightman et al. 2019; Wakefield et al. 2020a). By dividing gobbling activity into 3 stages, we were able to investigate effects of different predictor variables on variation in gobbling activity before, during, and after hunting and removal of males.

The final data set included gobbling activity at each ARU across days in each stage, distance to private property, distance to public access, distance to primary and secondary roads, distance to all land cover covariates, and mean predicted probability of use at each ARU by coyotes, male turkeys, and female turkeys (Table 1). We constructed 3 models to test our hypothesis and predictions regarding relative importance of the aforementioned variables on gobbling activity. Specifically, the first model was a predation risk model that included predicted probability of use by coyotes during each stage, along with metrics serving as proxies for hunting activity (distance to public access, distance to private property, and distance to secondary and primary roads) relative to gobbling activity. The second model focused on turkey resource selection and included predicted probability of use for males and females during each stage relative to gobbling activity. The third model was a land cover model that included distance to hardwoods, pine, openings, water/wetland, shrubs, and mixed pine-hardwoods, relative to gobbling activity.

Before analyses, we created a correlation matrix to test for multicollinearity among all covariates and for specific covariates in the approximating models, then removed covariates to avoid multicollinearity if r ≥ 0.60. We fitted each model as a negative binomial mixed model and used the negative binomial distribution as it best addressed over dispersion based on LOO validation (Harrison 2015; Vehtari et al. 2017). The response variable was the average daily gobbling count at each ARU, and all models included an ARU identification number as a random effect. To improve model fit and allow for direct comparison of effect sizes of each predictor variable, we normalized all fixed effects included in the models using the scale function in R (R Core Team 2021). We fitted models using Bayesian methods as implemented in Stan (Stan Development Team 2021) via the function brm from the package brms (Bürkner 2017) in R (R Core Team 2021). We computed 4 MCMC chains for 8,000 iterations, discarding the first 1,000 iterations as a burn-in. To guard against overfitting and help determine relative significance of variables within models, we used the lasso prior (Park and Casella 2008; van Erp et al. 2019). We assessed convergence of parameters by examining trace plots, posterior distributions, and calculating the Gelman-Rubin convergence diagnostic (values < 1.05 indicate convergence). We compared model fit by calculating LOO validation information criterion (LOOIC) and marginal/conditional R2 which indicates the amount of variation explained by fixed and random effects in the hierarchical models (Nakagawa and Schielzeth 2013).

Once the models were fit, to investigate whether each variable affected gobbling activity we calculated the probability of direction (pd), also known as the maximum probability of effect, using package bayestestR (Makowski et al. 2019a) with R (R Core Team 2021). Values vary between 0.5 and 1 and can be expressed as the probability that the variable of interest had a positive or negative effect on a parameter of interest (Makowski et al. 2019b). Using a pd value greater than 0.89 (Makowski et al. 2019b), we selected variables that influenced gobbling activity in each model to incorporate into a final candidate model to investigate holistically the effects of predation risk, turkey resource selection, and land cover type on gobbling activity. After fitting the final candidate model, we calculated the region of practical equivalence (ROPE) to determine significance of effects. The ROPE is the percentage of the high-density interval (HDI, ci = 0.89) of the posterior distribution that lies between -0.1 and 0.1, and ROPE values closer to 0% represent a significant effect (Kruschke 2018; Makowski et al. 2019b). Using a pd value greater than 0.89 and ROPE value less than 10%, we determined what variables in the final model had a significant effect on gobbling activity.

Results

Using 54 ARUs, we collected audio recordings across 12,325 days from 1 March through 30 June 2018 and 2019. The CNN selected 214,109 potential gobbles of which 53,025 (25%) were confirmed as a gobble; 45% were detected prior to hunting, 48% during hunting, and 7% after hunting. We noted variation in gobbling activity across ARUs before, during, and after hunting (Online Resource 2: Fig. 1). Mean (x̄ ± SE) number of gobbles per day across ARUs prior to hunting was 8 ± 1 and ranged from 0 to 62. Conversely, mean number of gobbles per day across ARUs during hunting was 5 ± 1 and ranged from 0 to 89, whereas after hunting gobbles declined to 1 ± 0.23 and ranged from 0 to 18 across ARUs. Mean predicted probability of use for turkeys and coyotes from RSF analysis along with summary statistics for all predictor variables across ARUs and stages can be found in Online Resource 2: Table 3.

The best approximating model from our 3 initial models was the predation risk model based on LOOIC and R2 values (Table 1). Across all 3 models, there was a significant effect of stage on predicted gobbling activity, wherein predicted gobbling activity was greatest prior to hunting, less during hunting, and least after hunting (Fig. 24, Online Resource 2: Table 4). Based on pd scores from the predation risk model, increased distance from public access during hunting had a positive effect on predicted gobbling activity whereas increased distance from private property had a negative effect, whereas predicted coyote risk had no effect (Fig. 2, Online Resource 2: Table 4). Based on pd scores from the turkey use model, we determined that after hunting, areas with greater predicted female use within home ranges were associated with greater predicted gobbling activity (Fig. 3, Online Resource 2: Table 5). Also, areas of increased predicted male use during the hunting season were associated with increased gobbling activity (Fig. 3, Online Resource 2: Table 5). Based on pd scores from the land cover model, we failed to determine any land cover types that had a positive or negative effect on predicted gobbling activity during any stages (Fig. 4, Online Resource 2: Table 6).

Fig. 2
figure 2

Probability of direction and magnitudes of conditional effects for each predictor variable (parameters) from Bayesian regression models estimating effects of predation risk on gobbling activity of male eastern wild turkeys on Cedar Creek and B.F. Grant wildlife management areas in Georgia, USA, during March–June 2018–2019. Each model includes an interactive term for stage [before (Pre-hunt), during (Hunt), and after (Post-hunt) hunting] and a random intercept term to account for variation across autonomous recording units

Fig. 3
figure 3

Probability of direction and magnitudes of conditional effects for each predictor variable (parameters) from Bayesian regression models estimating effects of 3rd order resource selection predicted probability of use by male and female eastern wild turkeys on gobbling activity on Cedar Creek and B.F. Grant wildlife management areas in Georgia, USA, during March–June 2018–2019. Each model includes an interactive term for stage [before (Pre-hunt), during (Hunt), and after (Post-hunt) hunting] and a random intercept term to account for variation across autonomous recording units

Fig. 4
figure 4

Probability of direction and magnitudes of conditional effects for each predictor variable (parameters) from Bayesian regression models estimating effects of land cover on gobbling activity of male eastern wild turkeys on Cedar Creek and B.F. Grant wildlife management areas in Georgia, USA, during March–June 2018–2019. Each model includes and interactive term for stage [before (Pre-hunt), during (Hunt), and after (Post-hunt) hunting] and a random intercept term to account for variation across autonomous recording units

Our final model based on variables with > 90% pd consisted of distance to public access, distance to private property, predicted third order male and female turkey probability of use; overall this model best fit the data based on LOOIC and R2 values (Table 1). Based on pd and ROPE values, stage and distance to public access during hunting were the only significant variables (Online Resource 2: Table 7). During hunting, distance to public access had a 98% probability of being positive and had strong evidence (0.10% in ROPE), with gobbling predicted to increase by a factor of 1.4 for every 500 m from public access (Fig. 5, Online Resource 2: Table 7). During hunting, distance to private land had a 94% probability of being negative but had weaker significance (12% in ROPE), where we would expect gobbling to decrease by a factor of 0.79 for every 500 m from private land (Fig. 6, Online Resource 2: Table 7). After hunting, predicted female probability of use had a 90% probability of being positive and during hunting, predicted male probability of use had a 91% probability of being positive. However, neither of these variables were found to be significant (17% in ROPE; Online Resource 2: Table 7).

Fig. 5
figure 5

The predicted effect with 95% credible intervals of distance to public access on average daily gobbling activity of male eastern wild turkeys during spring hunting season on B. F. Grant and Cedar Creek Wildlife Management Areas in Georgia, USA, 2018–2019

Fig. 6
figure 6

The predicted effect with 95% credible intervals of distance to private property on average daily gobbling activity of male eastern wild turkeys during spring hunting season on B. F. Grant and Cedar Creek Wildlife Management Areas in Georgia, USA, 2018–2019

Discussion

The ability for male wild turkeys to effectively and safely gobble to attract females, maintain dominance hierarchies, and secure breeding opportunities is linked to the likelihood of successfully reproducing and surviving (Emlen and Oring 1977; Lima 1998; Oseen 2002; Anderson and Simmons 2006). Therefore, males must identify areas on the landscape that provide opportunities for mate attraction with reduced predation risk. To our knowledge, our study was the first to investigate predation risk from both a natural and human predator on gobbling activity of male wild turkeys, and to ultimately investigate potential predictors of spatial variation in gobbling activity. We found that increased distance away from public access routes used by hunters during the hunting season was the only statistically significant and most important predictor of increased gobbling activity. We also noted that increased distance from private land (a proxy for hunting activity and pressure) negatively influenced gobbling activity. Although this results was not statistically significant, the strength of the relationship suggests it is likely biologically relevant. Our findings offer opportunities to identify and promote areas on the landscape where gobbling activity is predicted to be highest, which has implications to hunter satisfaction and breeding activities of wild turkeys (Casalena et al. 2011; Chamberlain et al. 2018; Wakefield et al. 2020a).

Our findings contribute to a growing body of literature demonstrating that game species alter behaviors in response to hunting to avoid real or perceived predation risk (Ordiz et al. 2012; Gross et al. 2015; McGrath et al. 2018; Mohlman et al. 2019; Wightman et al. 2019). Given that hunter activity is highly correlated with landscape features that facilitate ease of access, such as trails (Diefenbach et al. 2005; Karns et al. 2012; Gerrits et al. 2020), we were not surprised that predicted gobbling activity increased with increasing distance from public access areas. We offer that several plausible, non-mutually exclusive, explanations exist. First, hunting activity is known to prompt males of various species to move away from areas of greater hunting pressure, presumably to avoid risk (Gross et al. 2015; Little et al. 2016b). Further evidence for this explanation was supported by results from our male turkey RSF, where during and after hunting, males selected to be farther away from public access compared to before hunting (Online Resource 2: Table 2). Therefore, because males are selecting to be farther away from access points, gobbling activity is likely increased in these parts of the landscape. Second, males closer to areas readily accessible by hunters would presumably experience greater risk and reduced survival (Plante et al. 2017; Leclerc et al. 2019) as they presumably encounter hunters at a greater rate (Wakefield et al. 2020b). Third, hunters are less likely to travel greater distances from public access areas, so males occupying parts of the landscape farther from public access areas likely experience reduced risk (Diefenbach et al. 2005; Karns et al. 2012; Lebel et al. 2012; Gerrits et al. 2020). Therefore, the reward associated from gobbling for individuals farther from distance to public access may be greater than the risk, compared to individuals who are closer to such access points.

Our findings suggest that the distribution and abundance of private lands may be an important influence on gobbling activity on adjacent public lands. Individual mortality risk for hunted species is often greater on public than private lands, due to increased hunter densities and pressure on public lands (Root et al. 1988; Small et al. 1991; Haus et al. 2019; Palumbo et al. 2019). As a result, individuals on public lands may shift use to private property with the onset of hunting (Burcham et al. 1999; Proffitt et al. 2013), which also has been observed in wild turkeys (Gross et al. 2015). Likewise, previous studies have noted substantive differences in harvest rates of wild turkeys on public lands relative to private lands (Hubbard and Vangilder 2005). Movements of hunted individuals from public to private lands, coupled with reduced harvest rates on private lands, could logically result in greater abundance of turkeys and increased gobbling activity on private properties adjacent to heavily hunted public lands (Haus et al. 2019; Wightman et al. 2019; Wakefield et al. 2020b; Eliason 2021). Previous studies have documented the use of refuges/sanctuaries and/or hunting quotas to limit hunter densities and access, which can reduce effects of hunting activity and harvest on target species, while also facilitating a source population that supports harvest on heavily pressured areas (McCullough 1996; Margules and Pressey 2000; Baskett et al. 2005; Hansen and Defries 2007; Dunlop et al. 2009). Managers may consider such strategies on tracts of public land exposed to substantive hunting pressure, with the goal of improving gobbling activity by mimicking conditions found on private properties.

Previous studies have noted significant effects of natural predators on calling frequency and courtship behaviors in various species (Belwood and Morris 1987; Boyko et al. 2004; Coleman et al. 2014). The risk allocation hypothesis predicts that prey should exhibit strongest antipredator behaviors during brief high-risk situations within low background risk, and weakest behaviors during pulses of safety within high background risk (Lima and Bednekoff 1999; Creel et al. 2008). We failed to observe any biologically relevant effects of predicted coyote use on gobbling activity. We note that our assessment of coyote risk necessarily occurred at a relatively large spatial scale given our methodology, but there is potential coyote use would affect gobbling activity at smaller spatial scale, such as when turkeys and coyotes directly interact. Future studies should attempt to determine the effects of known encounters of predators and turkeys on gobbling activity. Collectively, our findings indicate that metrics associated with hunting activity are more impactful on gobbling than coyote use of the landscape. This finding aligns with the risk allocation hypothesis, as hunters are an ephemeral predator on the landscape for a relatively short period of time and represent high risk and increased lethality compared to coyotes (Relyea 2003; Morosinotto et al. 2010).

We observed that predicted gobbling activity was positively related to probability of use by females before and after hunting, although the finding was not statistically significant and we predicted that this would occur during all stages. However, previous studies have shown that hunting and the removal of males negatively impacts gobbling activity (Hoffman 1990; Chamberlain et al. 2018; Wightman et al. 2019) while also prompting surviving males to reduce gobbling activity as hunting season progresses (Wakefield et al. 2020a). Therefore, it’s plausible that males may increase gobbling activity in areas of greater female use as such behavior would represent a situation when the potential reward of gobbling (i.e., increased mating opportunities) may outweigh risk of predation (Healy 1992; Jennions et al. 1997; Zuk and Kolluru 1998). However, we observed a negative but non-significant effect of female use on gobbling activity during hunting, suggesting that even in areas of predicted increase in female use the reward associated with gobbling may not outweigh the risk while hunting is occurring.

We suspect that our initial prediction of increased gobbling activity in areas of increased male use was likely too simplistic, as wild turkeys and other similar species that use conspicuous vocalizations and displays typically select particular sites within their ranges for mate attraction (Healy 1992; Cole 2013; Brenner et al. 2019). Furthermore, most gobbling activity in wild turkeys occurs around sunrise when males are at or near their roost sites (Wightman et al. 2019; Wakefield et al. 2020a), and roost sites are not uniformly distributed within male home ranges (Gross et al. 2015; Indrani et al. 2018; Wakefield et al. 2020b). Therefore, the lack of significance of predicted male use on gobbling activity could be explained by discrepancies in areas used by males for gobbling versus other daily activities, such as foraging and loafing. Moreover, it may be advantageous for an individual to call from one location and only stay there for a limited amount of time (e.g., roost site), as calling attracts females but also predators (Magnhagen 1991).

This study uniquely coupled spatially explicit acoustic and movement data, and used landscape features to explain spatial variation in an auditory courtship behavior. Collectively, our findings suggest that metrics associated with hunting activity and hunter presence most influenced spatial variation in gobbling. Furthermore, given that gobbling activity is directly linked to both hunter satisfaction and turkey reproduction, and expected daily gobbles during the hunting season increased by 40% for every 500 m further from public access and 22% for every 500 m closer to private land, managers may wish to seek ways to reduce effects of hunting activity and hunter presence on gobbling activity. Our findings suggest selectively altering hunter access on the landscape (e.g., closing portions of roads to motor vehicles) to reduce the ease of hunter access to portions of public lands could promote increased gobbling activity. Similarly, managers could create areas of refuge or sanctuary on large tracts of public land that experience substantive hunting pressure, primarily to provide areas on the landscape where gobbling activity would not be influenced by hunting activity. Average home range size for male turkey’s range from 300–800 ha (Moore et al. 2008; Gross et al. 2015; Collier et al. 2017), the size of refuge and sanctuary areas should be commensurate with this spatial scale. We recommend further research to investigate how various anthropogenic activities such as hunting influence species behavior, reproductive success, and survival relative to dynamic landscape patterns and conditions.