Factors influencing productivity of eastern wild turkeys in northeastern South Dakota

Abstract Population growth is highly sensitive to changes in reproductive rates for many avian species. Understanding how reproductive rates are related to environmental conditions can give managers insight into factors contributing to population change. Harvest trends of eastern wild turkey in northeastern South Dakota suggest a decline in abundance. We investigated factors influencing reproductive success of this important game bird to identify potential factors contributing to the decline. We monitored nesting rate, nest survival, renesting rate, clutch size, hatchability, and poult survival of 116 eastern wild turkey hens using VHF radio transmitters during the springs and summers of 2017 and 2018. Heavier hens were more likely to attempt to nest than lighter hens, and adult hens were more likely to renest than yearling hens. Nest survival probability was lowest in agricultural fields relative to all other cover types and positively related to horizontal visual obstruction and distance to the nearest road. Daily nest survival probability demonstrated an interaction between temperature and precipitation, such that nest survival probability was lower on warm, wet days, but lowest on dry days. Egg predation was the leading cause of nest failure, followed by haying of the nest bowl and death of the incubating hen. Poults reared by adult hens had a greater probability of survival than poults reared by yearling hens. Our estimate of survival probability of poults raised by yearling hens was low relative to other studies, which may be contributing to the apparent regional population decline. However, there is little managers can do to influence poult survival in yearling hens. Alternatively, we found nest survival probability was lowest for nests initiated in agricultural fields. Wildlife‐friendly harvesting practices such as delayed haying or installation of flushing bars could help increase productivity of eastern wild turkey in northeastern South Dakota.

Although intrinsic and extrinsic environmental variables can be important determinants of turkey reproductive success, managers have little or no ability to influence these variables. In contrast, managers often have some ability to manipulate habitat-related environmental variables. One habitat-related environmental variable that can be important to turkey reproductive success is cover type.
For example, Clawson and Rotella (1998) found that artificial nests located in Conservation Reserve Program (CRP) fields had greater success relative to nests located in non-CRP cover types. In contrast, hens nesting in agricultural cover types such as alfalfa fields may be subject to increase risk of nest failure and hen mortality Shields & Flake, 2006;Vangilder & Kurzejeski, 1995).
South Dakota has seen large-scale landscape changes as grasslands have been converted into row crop or other agricultural cover types, with the greatest losses having occurred in the northeastern region (16.9% between 2006 and 2012) (Reitsma et al., 2014). This loss of grassland cover types parallels declines in the amount of land enrolled in CRP (Hellerstein, 2017), which has been particularly steep in northeastern South Dakota (USDA, 2016). Microhabitat conditions can also be strongly associated with reproductive rates. For example, increased visual obstruction can be positively associated with nest success (Badyaev, 1995;Lutz & Crawford, 1987), likely because of reduced detectability by predators.
For this study, we evaluated factors influencing reproductive success of turkeys in northeastern South Dakota. Harvest of turkeys in northeastern South Dakota during the spring prairie turkey season declined more than 50% between 2010 and 2016 (Huxoll, 2016), prompting managers to study potential causes of this apparent decline. Since productivity has a strong influence on population growth of this species Roberts et al., 1995;Rolley et al., 1998), understanding the factors that influence reproduction in this population is necessary to identify and potentially reverse the causes of this apparent decline. The objectives of this study are to

| Study area
The study was conducted in Codington, Deuel, Grant, and Roberts counties in northeastern South Dakota. The study area was split between the Minnesota River-Red River Lowland in the eastern half of the study area and Coteau des Prairies physiographic region in the western half (Flint, 1955;Johnson, Higgins, & Hubbard, 1995).
The Coteau begins in the northwest and extends in a southeasterly direction across the study area (Miller, Kempf, & Koopman, 1979).
On top of the Coteau, the relief is gently undulating to hilly, while down in the Lowlands the land is nearly level (Flint, 1955;Miller et al., 1979). Elevations ranged from over 600 m above mean sea level on top of the Coteau to about 300 m above sea level in the Lowland (Miller et al., 1979). Most of the study area consisted of privately owned lands with some state-owned (e.g., Game Production Areas) and federally owned (e.g., Waterfowl Production Areas) lands scattered throughout.
Northeastern South Dakota is in a humid continental climate region, with mean annual precipitation of 57 cm and mean annual temperature of 6.5°C across the study area (Menne et al., 2012).
Early spring snowfall is possible, with about one-quarter (26%) of the total annual snowfall occurring from March through May (Menne et al., 2012). About 60% of the total annual precipitation occurs during the nesting and brood-rearing seasons (April through August; Menne et al., 2012). Northeastern South Dakota received below average precipitation (i.e., rainfall) and approximately average temperatures during spring seasons (1 April-30 F I G U R E 1 Map of land cover types (adapted from the National Land Cover Database 2016 land cover raster layer; Yang et al. 2018) in Codington, Deuel, Grant, and Roberts Counties in northeastern South Dakota, USA June) over the course of this study (Figure 2). Spring and summer temperatures can be highly variable, with average minimum temperatures near 0°C in early spring to average maximum temperatures near 28°C during summer; however, normal mean temperatures for the spring and summer months range from 13 to 19°C (Menne et al., 2012).

| Capture and radio telemetry
We monitored reproduction of turkeys by fitting female turkeys with radio transmitters. We captured turkeys by first locating flocks of turkeys during the winter (1 January-31 March) and then baiting turkeys into capture sites. We captured turkeys using rocket nets (Thompson & Delong, 1967;Wunz, 1984). Following capture, we aged female turkeys as adult or yearling based on the presence or absence of barring on the 9th and 10th primary feathers (Williams, 1961) and weighed each bird. We secured 80-g very high-frequency (VHF) radio telemetry transmitters (Advanced Telemetry Systems) using a shock cord harness and backpack mount. Transmitters were <3% of the hens' body weight to reduce the risk of the transmitter interfering with survival and reproduction (Fair, Paul, & Jones, 2010). Transmitters were equipped with an activity signal, a nonmoving (loafing) signal that is activated instantaneously whenever the hen is not in motion, and mortality signal set to activate after 8 hr of inactivity. We monitored turkeys 6 days per week during the spring and summer (1 April-31 July) by locating each transmitter signal and listening to the nature of the signal; however, turkeys that were incubating nests were monitored daily. A moving signal indicated the hen was alive but not incubating a nest, a nonmoving signal indicated the hen was alive and incubating a nest, and a mortality signal indicated a hen was no longer alive. All handling, marking, and monitoring proce-

| Nest marking and monitoring
We monitored nesting activity of hens daily from 1 April to 6 August, 2017-2018. We first determined onset of incubation by listening for nonmoving signals from VHF transmitters. Once a nonmoving signal was obtained, we located nesting hens via homing and marked the nest. We marked the nest area with ~4 flags at distances of 20-40 m from the nest bowl depending on cover height and density of vegetation while attempting to minimize disturbance. If a nonmoving signal was observed on a subsequent day, we assumed that the hen was still tending the nest. If a moving signal was observed, we visually inspected the nest bowl to determine whether the hen was temporarily away (i.e., eggs and nest bowl still active and not disturbed), or whether it was lost due to predation (i.e., smashed or removed eggs).
If a mortality signal was observed, we assumed that the hen died while tending the nest and we located the transmitter and assessed the cause of death. We classified nests as successful by the presence of hatched eggshells, or as failed if nest contents were depredated, destroyed, or abandoned (Lehman et al., 2001). If the nest was successful, we determined the number of eggs that hatched from the total clutch size by counting eggshell fragments and membranes (Lehman et al., 2001). We counted the number of eggs in failed nests to determine clutch size if the eggs were relatively intact and undisturbed (Lehman et al., 2001). If the clutch size of failed nests could not be accurately determined, we did not include that nest in the analysis of clutch size.

| Poult monitoring
We determined the initial number of poults that hatched from each successful nest based on egg shell and membrane remains (Lehman et al., 2001). The number of poults in each brood was counted at 1, 2, and 4 weeks posthatch by observing broods feeding in open areas ; however, if dense vegetation interfered with observations, broods were flushed to count poults. Broods F I G U R E 2 Total precipitation accumulation (i.e., rainfall) (cm) and mean air temperature (°C) during the springs (1 April to 30 June) of 2017 and 2018 in northeastern South Dakota, USA. The 30-year average  for total precipitation accumulation (24.0-cm) and mean air temperature (13.1°C) during spring in Milbank, South Dakota, USA, are indicated by the horizontal dashed lines (Menne et al., 2012) often formed crèches (multiple hens with a group of commingled poults) after poults were 2 weeks old, and crèches were common when broods were 4 weeks old, making it difficult to differentiate individual broods during the day. If we could not determine the number of poults in a brood during the day due to the formation of a crèche, we performed another poult count for that brood again at night. During night brood counts, we observed the brood while in the roost with the hen, being careful to not flush the group from the roost, to obtain an accurate count of poults ).

| Environmental and spatial covariate estimation
We sought to determine how nest-site characteristics influenced nest success. Therefore, we quantified nest-site characteristics on the hatch date for successful nests and on the projected hatch date for failed nests (Gibson, Blomberg, & Sedinger, 2016;McConnell, Monroe, Burger, & Martin, 2017;Smith et al., 2018). We measured understory visual obstruction readings (VOR) of vegetation by placing a Robel pole with 2.54 cm increments in the nest bowl and at 1 m from the nest in the four cardinal directions (Benkobi, Uresk, Schenbeck, & King, 2000;Robel, Briggs, Dayton, & Hulbert, 1970).
We recorded the lowest visible increment on the pole from a distance of 4 m while kneeling to a height of 1 m Robel et al., 1970). We measured VOR from the four cardinal directions at the nest bowl; however, at the peripheral 1 m from the nest measurements, we estimated VOR from only three cardinal directions, ignoring the 4th direction back across the nest bowl so as not to duplicate visual obstruction readings across the nest bowl . We measured the height (in centimeters) of living vegetation at the nest bowl and at 1 m from the nest in the 4 cardinal directions . We estimated total cover of grass, forbs, shrubs, and other cover using a 0.1-m 2 quadrat at the nest bowl, and at 5, 1-m intervals in the cardinal directions (Daubenmire, 1959). We qualitatively categorized the dominant land cover type within the area surrounding the nest bowl as either grassland, pasture, agriculture, or forest. We classified the land cover as pasture if grazing was currently occurring or had occurred that year. Alfalfa hayfields and row crop fields were classified as agriculture. CRP grasslands, old fields, and other land cover where the dominant vegetation was grass and forbs and where grazing did not occur were classified as grasslands. If a nest was located within a road ditch, we classified the land cover according to the dominant land use adjacent to the road ditch (e.g., a nest placed in a road ditch next to a corn field would be classified as agriculture). We used ArcMap version 10.6.1 (Environmental Systems Research Institute) to calculate the distance from each nest to the nearest road (i.e., interstate, federal highway, state highway, local paved road, local unpaved road), obtained from the South Dakota Department of Transportation (SDDOT, 2017).
We placed 10 precipitation and temperature monitoring stations throughout the study area before the onset of nesting and retrieved the monitoring stations after all broods were >4 weeks old. Monitoring stations consisted of a rain gauge and a HOBO Pendant Temperature Data Logger (Onset Computer Corporation) that recorded 4 or 6 temperature readings at evenly spaced intervals each day. Monitoring stations were placed systematically throughout the study area to cover the extent of all radio-marked hen locations. Rain gauges were checked after every precipitation event, and we calculated daily precipitation accumulation (mm) (hereafter precipitation) and daily mean temperature (°C) (hereafter temperature) for each monitoring station for each day of the study. We assigned precipitation and temperature covariates to each individual nest and each individual brood (for nest and poult survival analyses, respectively) by assuming precipitation and temperature conditions at the location of each nest and at the location of each brood was equal to the precipitation and temperature conditions observed at the closest monitoring station.

| Nesting rate
We modeled nesting rate as the probability an individual hen that was alive on 1 April would attempt to nest that year using Bayesian logistic regression. We modeled nesting rate as a function of the age-class of each hen (adult or yearling), year of the study (2017 or 2018), and weight of each hen (kg). We used informative prior distributions for the intercept (log odds a juvenile initiates a nest when all other coefficients fixed at 0) and the slope coefficient describing the difference in log odds of nesting (i.e., log odds ratio [LOR]) between adult and juveniles. Drawing upon the studies in Table 2, we used a Gaussian (mean = 0.9; SD = 0.2) prior distribution for the intercept coefficient and a Gaussian (mean = 1.6; SD = 0.8) prior distribution for the adult LOR coefficient. Details on how we derived informative prior distributions are in Appendix 2. We assumed logistic (location = 0; scale = 1) prior distributions for all other slope coefficients.

| Nest survival
We assumed survival of nest i during day t was a Bernoulli random variable: where y it = 1 if nest i survived day t, y it = 0 if nest i failed during day t, and p it represents daily survival probability (Royle & Dorazio, 2008).
We further assumed a logit-linear model for daily survival probability which we model as a function of age-class of the nesting hen (adult or yearling), precipitation, temperature, land cover type (agriculture, forest, grassland, or pasture), mean VOR, mean total cover, and distance to the nearest road (m). We included an interactive effect of precipitation on temperature, since the effect of precipitation may vary depending on the temperature. Fifteen failed nests were missing VOR and total cover observations because the vegetation surrounding the nest was removed via haying before it could be measured. Rather than discard those nests, we imputed missing predictor variables, accounting for uncertainty in unmeasured variables (Gelman et al., 2013).
We accounted for repeated observations on individual nests by fitting a random coefficients model (Gelman & Hill, 2007). We assumed each coefficient β ji was a Gaussian random variable: where µ j and τ j represent the population-level mean and precision, respectively, of slope coefficient j. We used informative prior distributions for the intercept and adult LOR population-level mean parameters.
Drawing upon the studies in Table 2, we used a Gaussian (mean = 3.2; SD = 0.3) prior distribution for the intercept population-level mean parameter and a Gaussian (mean = 0.3; SD = 0.4) prior distribution for the adult LOR population-level mean parameter. Details on how we derived informative prior distributions are in Appendix 2. We selected logistic (location = 0; scale = 1) prior distributions for all other population-level mean parameters and gamma (shape = 1; rate = 1) prior distributions for precision parameters τ j , which provided little prior information.

| Renesting rate
We modeled renesting rate as the probability an individual hen would attempt a second nest, conditional upon failure of the first nest, using Bayesian logistic regression. We considered a hen unavailable for renesting if she was killed while tending the first nest.
We modeled renesting rate as a function of the age-class of each hen (adult or yearling), year of the study (2017 or 2018), ordinal date of failure for the previous nest attempt, length of the incubation period for the previous nest attempt (days), and weight of each hen (kg).
We used informative prior distributions for the intercept and adult LOR coefficients. Drawing upon the studies in Table 2, we used a Gaussian (mean = −0.7; SD = 0.6) prior distribution for the intercept coefficient and a Gaussian (mean = 0.6; SD = 0.7) prior distribution for the adult LOR coefficient. Details on how we derived informative prior distributions are in Appendix 2. We selected logistic (location = 0, scale = 1) prior distributions for all other slope coefficients.

| Clutch size
We modeled mean clutch size using Bayesian Poisson regression based on the number of eggs laid in each individual nest. Nests were excluded from the analysis if an accurate count of eggs could not be obtained (i.e., the nest was depredated and only some egg fragments remained). We modeled clutch size as a function of age-class of the nesting hen (adult or yearling), year of the study (2017 or 2018), weight of the nesting hen (kg), and nest attempt (first or second). We used informative prior distributions for the intercept (log expected count when all other slope coefficients equal 0) and the coefficient describing the difference in log expected count between adults and juveniles. Drawing upon the studies in Table 2, we used a Gaussian (mean = 2.4; SD = 0.4) prior distribution for the intercept coefficient and a Gaussian (mean = 0.0; SD = 0.6) prior distribution for the adult coefficient. Details on how we derived informative prior distributions are in Appendix 2. We selected Gaussian (mean = 0; SD = 1) prior distributions for all other slope coefficients.

| Hatchability
We modeled hatchability as the proportion of eggs that hatched from each individual nest based on the total number of eggs laid in each individual nest using Bayesian logistic regression. We modeled hatchability as a function of age-class of the nesting hen (adult or yearling), year of the study (2017 or 2018), and weight of the nesting hen (kg). We used informative prior distributions for the intercept and adult LOR coefficients. Drawing upon the studies in Table 2, we used a Gaussian (mean = 1.3; SD = 0.9) prior distribution for the intercept coefficient and a Gaussian (mean = −0.2; SD = 1.1) prior distribution for the adult LOR coefficient. Details on how we derived informative prior distributions are in Appendix 2. We selected logistic (location = 0; scale = 1) prior distributions for all other slope coefficients.

| Poult survival
We assumed the number of poults alive in each brood i at each day posthatch t was a Binomial random variable: where N i1 was equal to the initial number of poults in each brood i and φ it represents daily survival probability of each poult in brood i between time t − 1 and time t. If a brood-rearing hen died during the 28-day posthatch interval, we assumed N it = 0 for all subsequent poult counts. We treated the number of poults counted during each of 3 poult monitoring events at 7, 14, and 28 days posthatch (t = 8, 15, 29) as fixed and known, and treated N it as a latent random variable during all other time steps.
We further assumed a logit-linear model for daily survival probability which we model as a function of brood age (1-28 days posthatch), age of the brood-rearing hen (adult or yearling), year of the study (2017 or 2018), precipitation, and temperature. We additionally modeled the interaction between precipitation and temperature. We used an informative prior distribution for the intercept coefficient. Note that we did not include an informative prior distribution for the difference in log odds of poult survival between those raised by adults and juveniles because previous studies (Table 2) did not make this distinction. Drawing upon the studies in Table 2, we used a Gaussian (mean = 3.4; SD = 0.1) prior distribution for the intercept coefficient. Details on how we derived informative prior distributions are in Appendix 2. We selected logistic (location = 0; scale = 1) prior distributions for all other slope coefficients.
We fit all models using Bayesian methods to maintain a consistent analytical approach. We fit each model with JAGS version 4.3.0 (Plummer, 2003) via the jagsUI version 1.4.9 interface (Kellner, 2018) in program R version 3.5.1 (R Core Team, 2018). We ran three chains for each model using trace plots to determine an adequate burn-in period and subsequently ran models until we achieved reasonable convergence (R ≤ 1.1; Gelman et al., 2013). We concluded that slope coefficients were different from 0 if 95% credible intervals (CI) did not overlap 0.   Table A1). Nesting rate did not differ between years of the study (Table A2).

| Nest survival
We observed a total of 147 nest attempts during this study. In 2017, 48 hens attempted one nest, seven hens attempted two nests, and one hen attempted three nests (65 nests total). In 2018, 42 hens attempted one nest, 17 hens attempted two nests, and two hens attempted three nests (82 nests total). We recorded at least one nest attempt in both years from 28 hens. Across both years of the study, five nests were censored due to investigator interference causing the hen to abandon the nest. Five additional nests were censored because we were unable to visit the nest site due to a lack of landowner permissions. Therefore, nest survival probabilities were estimated from a total of 137 nests, across 2,412 days where an individual nest was at risk of failure.
Fifty-six of 137 nest attempts were successful. Predation of eggs was the leading cause of nest failure, accounting for over half of all failed nest attempts ( Table 1). Haying of vegetation surrounding the nest and death of the incubating hen were also major sources of nest failure, accounting for 16% and 12%, respectively, of all failed nest attempts (Table 1). Over a 28-day average incubation period, Death of incubating hen a 10 12% Haying 13 16% Predation of eggs 47 58% Trampled by livestock 4 5% Total 81 a Nine events caused by predation; 1 event caused by a vehicle collision lower success probability relative to nest located in any other cover type ( Figure 4, Table A2). All nests that were laid in alfalfa fields failed because fields were hayed (n = 10) or depredated (n = 2) before eggs hatched. We also found nests with greater visual obstruction ( Figure 5, Table A2) and that were placed further from roads ( Figure 6, Table A2) had higher daily nest survival probability. Finally, precipitation and temperature had an interactive effect on daily nest survival probability. On relatively cool days, we found a positive effect of precipitation on daily nest survival probability. However, as temperature increased, the effect of precipitation on daily survival diminished. Thus, for a fixed amount of precipitation, predicted daily nest survival tended to be lower on warmer days (Figure 7, Table A2).
Daily nest survival probability was not strongly affected by mean total cover (i.e., 95% CI of the population-level mean overlapped 0; Table A2).

| Renesting rate
Renesting rate probabilities were estimated from a total of 58 hens that were available to renest after a failed first nest attempt ( Table A3). Renesting rate did not differ between years of the study and was not affected by hen weight at the time of capture or duration of previous nesting attempt (Table A3). accurate clutch counts for 40 out of 91 failed nest attempts and two out of 56 successful nest attempts over the course of this study, and therefore, omitted these nests from the clutch size analysis. During the first nest attempt, the mean clutch size laid by adult hens was 10.6 (95% CI = [9.8, 11.5]), the mean clutch size laid by yearling hens was 10.0 (95% CI = [8.9, 11.2]), and clutch size did not vary by ageclass. Clutch size did not differ between nesting attempts or years of the study and was not affected by hen body weight at the time of capture (Table A4). increased (i.e., as poults become older; Figure 9, Table A6). Daily poult survival probability was not affected by precipitation or temperature and did not differ between years of the study (Table A6).

| D ISCUSS I ON
Our investigation of a turkey population exhibiting an apparent decline allowed us to identify important sources of variation in reproductive rates and identify factors potentially contributing to the apparent decline. Many of our estimates of reproductive rates-nest survival, renesting rate, clutch size, and hatchability-were well within the range of previously reported estimates for these reproductive rates (Table 2). However, adult hen nesting rate and the probability of poults raised by yearlings surviving to 28 days were low when compared to other studies (Lehman et al., 2001;Shields & Flake, 2006) (Table 2). Poult survival to 28 days posthatch is an important demographic rate affecting turkey population dynamics (Hubbard, Garner, & Klaas, 1999;Vangilder & Kurzejeski, 1995), and the low survival probability of poults raised by yearlings could be contributing to decreased productivity of hens in northeastern South Dakota and the apparent population decline.
Although we found nest survival probability was well within the range of published estimates, we identified important sources of variation in this reproductive rate. In particular, we found that nest survival was substantially lower in agricultural fields relative to other cover types. This was largely driven by failure of all nests placed in alfalfa fields, which were mostly lost due to haying operations. Haying can be an important cause of mortality and nest failure for turkey and other avian species within agricultural landscapes (Bollinger, Bollinger, & Gavin, 1990;Shields & Flake, 2006;Wright, Paisley, & Kubisiak, 1996). Standard recommendations for wildlife-friendly farming include delaying haying until mid-July to allow nests in agricultural fields to fledge, haying from the center of the field and working out to allow nesting birds to escape, and installing flushing bars to minimize mortality (NRCS, 1999). Incentive programs designed to restrict haying during grassland bird nesting seasons have been demonstrated to improve avian reproductive success (Perlut, Strong, & Alexander, 2011). However, delaying haying operations can come at high economic costs to farmers and such wildlife-friendly practices are unlikely to succeed without established incentive programs.
Like the agricultural activities describe above, predation is another important source of nest failure. Nest-site selection is strongly driven by avoidance of predation (Gill, 2007). Turkeys are known to select nests sites that are visually obstructed to predators (Isabelle, Conway, Comer, Calkins, & Hardin, 2016;Wood, Cohen, Conner, Collier, & Chamberlain, 2019), and our study demonstrated that selection for nest sites with high visual obstruction may reduce predation risk, as the probability of nest success was greater at sites with relatively high visual obstruction (Yarnall, Litt, & Lehman, 2019).
Turkeys may also select nests located farther away from roads as a predator avoidance strategy. Beasley, DeVault, and Rhodes (2007) noted high raccoon prevalence near roads, and coyotes are known to use roads as travel corridors and forage along roads (Tigas, Van Vuren, & Sauvajot, 2002;Gosselink, Van Deelen, Warner & Joselyn, 2003;Hinton, van Manen, & Chamberlain, 2015). Our study found that nests located close to roads had a lower probability of survival, likely due to increased predation risk near roads. This result is similar to Badyaev (1995), who found that successful turkey nests were located farther from roads, on average, than unsuccessful nests.
While many predators hunt visually, many also hunt by olfaction. Roberts and Porter (1998b) and  found a negative association between precipitation and nest survival probability, presumably because of increased detection of incubating hens by predators. Our finding of a positive association with daily nest survival probability and amount of precipitation at cold temperatures seemingly contradicts these findings. However, we found that for a given amount of precipitation, survival probability decreased as temperature increased, which is consistent with Roberts and Porter's (1998b) suggestion that olfactory cues to locate nests may be enhanced during warm, wet periods. Alternatively, the nesting seasons captured during this study were, on average, drier than a typical spring in our study area ( Figure 2). It is possible that the effects of precipitation on nest survival may extend beyond individual rainfall events and be cumulative in nature, such that rainfall events in relatively dry years may influence nest survival differently than rainfall events in relatively wet years.
In addition to external factors such as cover type, temperature, and precipitation, we found intrinsic factors such as experience and body condition were important sources of variation in reproductive rates. For example, nesting rate increased as hen body weight increased, likely because heavier birds are better able to complete the energetically expensive task of incubating a clutch (Porter et al., 1983). Following nest failure, adult hens were more likely to renest than yearling hens, which is consistent with previous research (Lehman et al., 2001;Shields & Flake, 2006). Similarly, poults reared by adult hens had a greater daily survival than poults reared by yearling hens, consistent with studies of Merriam's turkey in the Black Hills, South Dakota . Finally, poult survival probability increased with the number of days posthatch, which is also consistent with many other studies (Hubbard et al., 1999;Lehman et al., 2001;Paisley et al., 1998;Porter et al., 1983;Shields & Flake, 2006;Switzer & Tucker, 2009;Vander Haegen et al., 1988;Vangilder & Kurzejeski, 1995). Older poults are better able to escape ground predators (Spears et al., 2007) and thermoregulate (Healy & Nenno, 1985;Schmidt-Nielsen, 1997), increasing daily survival probability as they age.
An important source of uncertainty in the apparent regional decline in turkey abundance is the role of loss of land enrolled in the Conservation Reserve Program (CRP). The amount of land enrolled in CRP has been steadily declining for more than a decade (Hellerstein, 2017), with some of the steepest declines occurring in northeastern South Dakota (USDA, 2016). It is possible that conversion of former CRP lands to agricultural production could translate to increased probability of nesting in agricultural fields, leading to decreases in reproductive output and consequently declining population growth rates. Future research should determine how nestsite selection varies as a consequence of loss of CRP habitat, and whether loss of CRP habitat increases the probability of nesting in low productivity habitats such as agricultural fields.
Our research highlighted reproductive parameters that may be contributing to an apparent population decline in turkey in north- in other studies and there may be room to increase nest survival and reproductive output. In particular, we found low daily survival probability in agricultural fields relative to other cover types. To increase nest survival in these cover types, managers could promote wildlife-friendly practices such as delayed haying (NRCS, 1999).
Alternatively, increasing the availability of suitable nesting cover types (e.g., CRP fields) may lead to lower probabilities of nesting in agricultural fields. Improvements to productivity and recruitment could potentially stabilize, or lead to growth of, the population of turkeys in northeastern South Dakota.

CO N FLI C T O F I NTE R E S T
The authors have declared that no competing interests exist.

O PEN R E S E A RCH BA D G E S
This article has earned an Open Data Badge for making publicly available the digitally-shareable data necessary to reproduce the reported results. The data is available at https://doi.org/10.5061/ dryad.5dv41 ns3d.

A PPEN D I X 2
Details regarding how informative prior distributions were obtained.

N E S TI N G R ATE , R E N E S TI N G R ATE , A N D H ATCH A B I LIT Y
We derived informative prior distributions for the intercept and adult log odds ratio (LOR) coefficients by drawing on those studies in Table 2 that distinguish between adult and juvenile nesting rate, renesting rate, and hatchability. For each reproductive parameter, we integrated previous studies into a beta-binomial model: where y i is the number of "successes" from study i (defined as hens nesting, hens renesting, or eggs hatching), N i is the total number of "trials" included in study i (hens available to nest, hens available to renest, or eggs laid), p i is the probability of success, and a and b are shape parameters governing the beta distribution. We assumed gamma where s a and s j are posterior distributions of s obtained from adults and juveniles, respectively.
Distributions of both s and t were very well approximated with a Gaussian distribution. We therefore used Gaussian distributions as prior distributions for the intercept and adult LOR coefficients. We used a Gaussian prior distribution with mean and SD calculated from posterior samples of s j for the intercept coefficient of our logit-linear models of nesting probability, renesting probability, and hatchability.
We used a Gaussian prior distribution with mean and SD calculated from posterior samples of t for the adult LOR slope coefficient of our logit-linear models of nesting probability, renesting probability, and hatchability.

N E S T S U RV I VA L
We derived informative prior distributions for the intercept and adult log odds ratio (LOR) coefficients by drawing on those studies in Table 2 that distinguish between adult and juvenile nest success.
We assumed the probability that a hen successfully hatched at least 1 poult reported from study i was a beta random variable: where a and b are shape parameters governing the beta distribution.
We assumed gamma (shape = 0.1; rate = 0.1) prior distributions for both a and b. We separately estimated posterior distributions of a and b for both adults and juveniles.
After obtaining posterior distributions of a and b for both adults and juveniles, we estimated the posterior distribution of a hen successfully hatching at least 1 poult as: which is the expected value of a beta random variable. Since we modeled the daily probability of nest survival on the logit-linear scale, we transformed the quantity r (which is bound between 0 and 1) to the log odds scale: which is the well-known logit transformation. Note that we also transformed the probability of hatching at least 1 poult to daily nest survival probability by taking the 28th root of r, to reflect a 28-day incubation period. To approximate the adult LOR coefficient, we calculated the difference in s obtained from juvenile and adult log odds of daily nest survival: where s a and s j are posterior distributions of s obtained from adults and juveniles, respectively.
Distributions of both s and t were very well approximated with a Gaussian distribution. We therefore used Gaussian distributions as prior distributions for the intercept and adult LOR coefficients. We used a Gaussian prior distribution with mean and SD calculated from posterior samples of s j for the intercept coefficient of our logit-linear models of daily nest survival. We used a Gaussian prior distribution with mean and SD calculated from posterior samples of t for the adult LOR slope coefficient of our logit-linear models of daily nest survival.

P O U LT S U RV I VA L
We derived informative prior distributions for the intercept coefficient by drawing on those studies in Table 2 that report the probability of poult survival to 28 days. We assumed the probability that a poult survived to 28 days reported from study i was a beta random variable: where a and b are shape parameters governing the beta distribution. We assumed gamma (shape = 0.1; rate = 0.1) prior distributions for both a and b. We did not distinguish between poults raised by adults or yearlings, as this distinction was not made in studies listed in Table 2.
After obtaining posterior distributions of a and b, we estimated the posterior distribution of a poult surviving to 28 days as: which is the expected value of a beta random variable. Since we modeled the daily probability of poult survival on the logit-linear scale, we transformed the quantity r (which is bound between 0 and 1) to the log odds scale: which is the well-known logit transformation. Note that we also transformed the probability of a poult surviving to 28 days to daily poult survival probability by taking the 28th root of r. The distribution of s was very well approximated with a Gaussian distribution. We therefore used a Gaussian prior distribution with mean and SD calculated from posterior samples of s for the intercept coefficient of our logit-linear model of daily poult survival.

CLUTCH S IZE
We derived informative prior distributions for the intercept and difference in log expected count between adults and juveniles (hereafter simply "adult") coefficients by drawing on those studies in Table 2 that distinguish between adult and juvenile clutch size. We assume mean clutch size reported from study i was a gamma random variable: where a and b are shape and rate parameters, respectively, governing the gamma distribution. We assumed gamma (shape = 1, rate = 1) prior distributions for both a and b. We separately estimated posterior distributions of a and b for both adults and juveniles.
After obtaining posterior distributions of a and b for both adults and juveniles, we estimated the posterior distribution of expected clutch size as: which is the expected value of a gamma random variable. Since we modeled expected clutch size on the log scale, we transformed the quantity r as: To approximate the adult coefficient, we calculated the difference in s obtained from juvenile and adult clutch size models: where s a and s j are posterior distributions of s obtained from adults and juveniles, respectively.
Distributions of both s and t were very well approximated with a Gaussian distribution. We therefore used Gaussian distributions as prior distributions for the intercept and adult coefficients. We used a Gaussian prior distribution with mean and SD calculated from posterior samples of s j for the intercept coefficient of our log-linear model of clutch size. We used a Gaussian prior distribution with mean and SD calculated from posterior samples of t for the adult coefficient of our log-linear model of clutch size.