Behavioral modifications lead to disparate demographic consequences in two sympatric species

Abstract Life‐history theory suggests species that typically have a large number of offspring and high adult mortality may make decisions that benefit offspring survival in exchange for increased adult risks. Such behavioral adaptations are essential to understanding how demographic performance is linked to habitat selection during this important life‐history stage. Though studies have illustrated negative fitness consequences to attendant adults or potential fitness benefits to associated offspring because of adaptive habitat selection during brood rearing, equivocal relationships could arise if both aspects of this reproductive trade‐off are not assessed simultaneously. To better understand how adaptive habitat selection during brood rearing influences demographics, we studied the brood survival, attendant parental survival, and space use of two sympatric ground‐nesting bird species, the northern bobwhite (hereafter: “bobwhite”; Colinus virgininanus) and scaled quail (Callipepla squamata). During the 2013–2014 breeding seasons, we estimated habitat suitability across two grains (2 m and 30 m) for both species and determined how adult space use of these areas influenced individual chick survival and parental risk. We found the proportion of a brood's home range containing highly suitable areas significantly increased bobwhite chick survival (β = 0.02, SE = 0.006). Additionally, adult weekly survival for bobwhite was greater for individuals not actively brooding offspring (0.9716, SE = 0.0054) as compared to brooding adults (0.8928, SE = 0.0006). Conversely, brood habitat suitability did not influence scaled quail chick survival during our study, nor did we detect a survival cost for adults that were actively brooding offspring. Our research illustrates the importance of understanding life‐history strategies and how they might influence relationships between adaptive habitat selection and demographic parameters.

neously. To better understand how adaptive habitat selection during brood rearing influences demographics, we studied the brood survival, attendant parental survival, and space use of two sympatric ground-nesting bird species, the northern bobwhite (hereafter: "bobwhite"; Colinus virgininanus) and scaled quail (Callipepla squamata).
During the 2013-2014 breeding seasons, we estimated habitat suitability across two grains (2 m and 30 m) for both species and determined how adult space use of these areas influenced individual chick survival and parental risk. We found the proportion of a brood's home range containing highly suitable areas significantly increased bobwhite chick survival (β = 0.02, SE = 0.006). Additionally, adult weekly survival for bobwhite was greater for individuals not actively brooding offspring (0.9716, SE = 0.0054) as compared to brooding adults (0.8928, SE = 0.0006). Conversely, brood habitat suitability did not influence scaled quail chick survival during our study, nor did we detect a survival cost for adults that were actively brooding offspring.
Our research illustrates the importance of understanding life-history strategies and how they might influence relationships between adaptive habitat selection and demographic parameters.

K E Y W O R D S
brood, Callipepla squamata, Colinus virginianus, demography, habitat suitability index, lifehistory strategy

| INTRODUC TI ON
Animal space use is a fundamental pattern in wildlife ecology that constitutes important mechanisms used to conserve and manage species worldwide (Lack, 1933;Manly, McDonald, Thomas, McDonald, & Erickson, 2002;Morrison, Marcot, & Mannan, 2012).
However, using this pattern to manage species can be misguided if the assumption that individuals are making selection decisions that maximize their demographic parameters (i.e., survival and/or fecundity) is incomplete. Research has often shown that space use can be misleading indicators of habitat quality (i.e., ecological traps; Bacon et al., 2016;Donovan & Thompson, 2001;Gates & Gysel, 1978;van Horne, 1983) because individuals may not always be able to determine habitat quality and instead rely on environmental cues to guide space use (Kristan, 2003;Storch & Frynta, 1999). Furthermore, this relationship can be highly variable across species and vegetation communities (Bock & Jones, 2004). Thus, determining a link between space use and demographic parameters helps ensure the predictive capabilities of using such patterns for conservation purposes (Beerens, Frederick, Noonburg, & Gawlik, 2015;Folmer & Piersma, 2012).
Further constrained within patterns of habitat selection is an understanding of how adult space-use patterns influence offspring survival. This is because the patterns of space use for offspring of many species (i.e., broods and neonates) are often ascribed to parental decisions of nest/birth site selection and subsequent habitat use during important life-history stages for offspring such as the brooding period (Dreitz, 2009;Gibson, Blomberg, Atamian, & Sedinger, 2017;Kolbe & Janzen, 2002;Lengyel, 2006). Such parental decisions play a direct role in influencing offspring survival (Dreitz, 2009;Garrick, Amundson, & Seddon, 2017;Gibson et al., 2017;Kolbe & Janzen, 2002), which in turn can directly affect population dynamics, as this life-history stage may be a critical period in certain species (Colwell, Hurley, Hall, & Dinsmore, 2007;Sandercock, Jensen, Williams, & Applegate, 2008).
Evidence suggests that for species with a large number of offspring and low adult survival (i.e., r-selected species), adult decision-making should reflect benefits toward offspring survival in exchange for increased adult risks, whereas the opposite tends to be true for species that have longer-lived adults and fewer offspring (Ghalambor & Martin, 2001). Such increased risks can be associated with physiological constraints (i.e., changes in metabolic requirements during brood rearing [Dawson, Hinsley, Ferns, Bonser, & Eccleston, 2000;Dreitz, 2009]), behavioral changes (i.e., increased vigilance, decreased foraging opportunities, and space-use change [Williams & Cooke, 1994;Zhao et al., 2018]), and/or changes in biotic interactions such as increased predation risk due to novel behaviors associated with an adult with offspring (i.e., feigning injuries [Bellrose & Holm, 1994;Ghalambor & Martin, 2001]). Thus, adult space use during these periods may be representative of environmental conditions that are more suitable toward offspring survival as opposed to parental survival. These patterns could convolute how models of habitat use are used in conservation planning and wildlife management if demographic consequences of such behavioral modifications are not considered.
Previous research exploring the links between habitat selection during the postnesting period and offspring survival has been inhibited due to logistical constraints and thus has been lacking until recently (Bock & Jones, 2004). Moreover, a dearth of knowledge exists for precocial species, likely due to past logistical constraints associated with highly mobile offspring (Bloom, Clark, Howerter, & Armstrong, 2013;Bock & Jones, 2004;Orange et al., 2016), and the studies that have investigated these species have been somewhat ambiguous in linking habitat selection and offspring survival (Aldridge & Boyce, 2007;Bloom et al., 2013;Dreitz, 2009;Gibson et al., 2017;Gregg & Crawford, 2009;Mathews, Tyre, Taylor, Lusk, & Powell, 2011). Similarly, though many studies illustrate fitness consequences for adults associated with adaptive behavior during the brooding life-history stage (Blomberg et al., 2013;Hagen, Pitman, Sandercock, Robel, & Applegate, 2011;Mangelinckx, Davis, Allen, Sullivan, & Blomberg, 2018;Zhao et al., 2018), rarely have studies linked adaptive brooding behaviors to offspring survival while simultaneously assessing fitness consequences of the attending adults. By decoupling these two demographic consequences, equivocal results may arise when attempting to link adaptive habitat selection to either offspring survival or attending adult survival separately because this pattern in space use may be more beneficial toward an unobserved demographic parameter rather than the observed parameter (Uboni, Smith, Stahler, & Vucetich, 2017). Thus, attempting to understand concurrent demographic trade-offs between offspring survival and adult survival associated with brood habitat selection is important when assessing the conservation implications of these behavioral modifications.
Using radiotelemetry data of individual chicks and brooding adults, we sought to determine whether multiscale habitat suitability indices (HSIs; Bacon et al., 2016;Guisan & Thuiller, 2005) Figure 1). Furthermore, we sought to determine whether such behavioral modifications by adults resulted in an increased risk for the associated adults by assessing weekly survival probabilities. We chose these two ground-nesting species as they occur sympatrically in areas of the Great Plains, USA, allowing for a direct comparison on how relative variations in life-history strategies  and habitat use during the breeding season  influence the link between demographic parameters and behavioral adaptations. The study of sympatric and closely related species may help to better describe the habitat-demographics link and identify factors that may limit population growth and viability of these species (Ackerman, Herzog, Takekawa, & Hartman, 2014;Koons & Rotella, 2003;Sieving, 1992;Varo, 2008). Previous research has indicated divergent reproductive strategies  within areas of sympatry between these two species (i.e., scaled quail having higher adult survival though generally producing less offspring compared to bobwhite [Rollins, 2000] During our study period (summers of 2013-2014), annual precipitation ranged from 394 mm in 2014 to 503 mm in 2013, with both years having drier conditions than the long-term  average of 560 mm for this area (Brock et al.., 1995;McPherson et al., 2007). Furthermore, average summer (May-July) temperatures ranged from 25.7 to 27.2°C in 2013 and 25.3 to 30.1°C in 2014, generally exceeding the long-term average summer temperature of 25.3°C for this region (Brock et al., 1995;McPherson et al., 2007

| Bird capture and monitoring
Bobwhite and scaled quail adults were captured using baited walk-in funnel traps and were fitted with 7-g necklace-style VHF radio-transmitters (Advanced Telemetry Solutions, Isanti, MN).
Nests were located via radiotelemetry and monitored daily after they were initially found. After hatching, broods were monitored daily via the radio-marked adult until the chicks reached 8-12 days old when they were captured for attachment of radio-transmitters. We used a combination of methods (Andes et al., 2012;Smith et al., 2003) to capture chicks. Following capture, chicks were held in a small portable cooler with a warm water bottle to prevent hypothermia. We attached transmitters to 50%-66% of chicks captured in each brood. Chicks were fitted with 0.45-g suturestyle transmitters that had an expected battery life of 21-23 days (American Wildlife Enterprises, Monticello, FL). Transmitters were attached using methods described by Burkepile, Conelly, Stanley, and Reese (2002) and Dreitz, Baeten, Davis, and Riordan (2011).
Attachment and capture protocols were approved by Oklahoma State University's Institutional Animal Care and Use Committee (ACUP #AG132 and #AG11-22). We acknowledge that a limitation of our study is that some individual chicks likely experienced mortality before our 8-to 12-day-old capture period, and thus, all survival estimates and space-use relationships exclude this initial period of high posthatch mortality (Terhune, Palmer, & Wellendorf, 2019).
Following capture, chicks were located daily. We located broods by homing (White & Garrott, 1990) to a distance of 15-20 m to minimize disturbance. If all radio-tagged chicks in a brood died, radiotagged brooding adults were flushed on a weekly basis to verify the presence of at least one chick with a brooding adult to estimate habitat use and movement. To minimize the effect of variation in diurnal habitat selection that can influence habitat analysis (Taylor, Church, & Rusch, 1999;Taylor & Guthery, 1994), we alternated collection of daily telemetry locations between two time periods: active (sunrise-11:00 and 16:01-sunset) and loafing (11:01-16:00).
Daily brood locations were used to calculate a relative index of average daily movement for broods . We estimated this index by using the Euclidean distance between daily consecutive brood locations (Brøseth & Pedersen, 2010;Lohr, Collins, Williams, & Castelli, 2011)

| Habitat suitability analysis for broods
To determine habitat suitability on our study site, we used a maximum entropy algorithm, Maxent version 3.3.3 (Phillips & Dudík, 2008). Traditionally, this algorithm has been used for species distribution models or ecological niche models (Elith et al., 2011).
However, Maxent may also be useful in determining smaller scale patterns of space use or habitat selection (Baasch, Tyre, Millspaugh, Hygnstrom, & Vercauteren, 2010;. We integrated the radiotelemetry locations from our broods as the occurrence dataset for our modeling exercise. Any identical occurrence locations (i.e., multiple locations within the same pixel of our environmental layers) were removed from the dataset prior to running models. Furthermore, we eliminated any broods with occurrence locations or home ranges located outside the extent of our study area as determined by the extent of our environmental layers used for modeling.

| Environmental layers
Similar to  and Tanner, Papeş, et al. (2017), environmental layers used for our Maxent modeling procedure represented the configuration and structure of vegetation within our study site. An initial vegetation layer was created using an Iso Cluster Unsupervised Classification method from 2-m resolution satellite imagery which was collected in July of 2013. This exercise resulted in a vegetation layer consisting of 10 primary cover types: mixed shrub, sand sagebrush, mixed grass, shortgrass/yucca, sparse vegetation, bare ground, salt cedar, open water, developed areas, and agriculture/food plots. Descriptions of these 10 primary cover types are given in Appendix S1.
To incorporate variability in resource selection across multiple scales, we used environmental layers at both 2 m and 30 m grains (discussed here) and across two extents: the buffered home range and study site extents (discussed in the next section). We incorporated both changes in grain and changes in extent to meet the qualifications of a multiscale study as described by McGarigal, Wan, Zeller, Timm, and Cushman (2016). To incorporate a change in grain, we used Block Statistics and the Resample tools in ArcGIS 10.2 (ESRI, 2011) with a majority rule for classification.
Following the creation of our vegetation layer, we calculated a normalized difference vegetation index (NDVI). Furthermore, we used FRAGSTATS 4.2.1.603 (McGarigal, Cushman, & Ene, 2012) and the vegetation layer to create class-and landscape-level metrics to incorporate as environmental layers for Maxent models.
FRAGSTATS metrics were calculated using a round moving window with a radius of 215 m, which was equivalent to the maximum average daily movement of broods across both species during our study.
To limit any chance of correlation or redundancy in metrics included in our analysis (Ritters et al., 1995), we selected FRAGSTATS metrics from our study area that were previously shown to be influential in space-use analyses for bobwhite during the breeding season Tanner, Papeş, et al., 2017). We also excluded any metrics that were highly sensitive to change when incorporating a change in grain size (Lustig, Stouffer, Roigé, & Worner, 2015).
Based on these criteria, we included 14 metrics: the coefficient of variation in patch size for mixed shrub, total landscape edge density (m/ha), edge density of specific vegetation types (mixed shrub, sand sagebrush, salt cedar, bare ground, and shortgrass/yucca), mean area of all vegetation patches (m 2 ), mean area bare ground and sand sagebrush patches (m 2 ), perimeter-area fractal dimension (i.e., shape complexity across all patches), perimeter-area fractal dimension of mixed shrub and sand sagebrush patches, and the contagion index.
The contagion index is a measure of interspersion of patch types and the overall patch dispersion, such that it is based on the probability of finding a pixel of type i adjacent to a pixel of type j (O'Neill et al., 1988). Thus, a value of 0 represents a landscape where every pixel is a different patch type and is maximally interspersed, and a value of 100 represents a landscape where all patch types are maximally aggregated (McGarigal et al., 2012). For all layers, we reclassified "no data" cells within the extent of our study site to 0 before incorporating them into the Maxent algorithm (Foley, Rueda, Peterson, & Wilkerson, 2008).
We also included distance-based variables that included the Euclidean distance (m) to anthropogenic features including oil/gas structures, artificial surface water sources, and four different types of roads (county road, primary WMA road, restricted access WMA road [truck and all-terrain vehicle {ATV} access], and restricted access WMA road [ATV traffic only]). We separated roads into four categories to represent varying levels of potential human disturbance (i.e., road traffic), as differences in bobwhite hazard rates have been indicated based on classified road types on our study site . We only included functioning artificial water sources (windmills with water tanks, gallinaceous guzzlers, and solar water wells that were providing surface water) in our analysis as they have been shown to influence bobwhite and scaled quail space use on our study site . We did not differentiate between functioning and nonfunctioning oil/gas wells to create our distance raster as only 6% of these wells were considered nonfunctioning . Spatial oil/gas well data were obtained from the Oklahoma Corporation Commission in 2013, county road data were obtained from the Oklahoma Department of Transportation (http:// okmaps.org/ogi/search.aspx), and all other road data were mapped via ground-truthing efforts. The vegetation, NDVI, distance-based, and FRAGSTATS layers were included as environmental layers in our initial modeling exercise described in the next section.

| Second-order Maxent modeling
We used a two-stage approach to model habitat suitability for both bobwhite and scaled quail through Maxent. These two stages included an analysis for second-order selection (i.e., individual home ranges) and an analysis for the entire population across our study site. Because we had repeated sampling of individuals (radiotelemetry) which were used for the occurrence (or presence) dataset for Maxent models, we initially built Maxent models for individual broods within their buffered home ranges (Baldwin, 2009). Brood home ranges were calculated using a 95% fixed-kernel method with least squares cross-validation within the Geospatial Modelling Environment 0.7.2.1 (Beyer, 2012) and were buffered by the species' specific average daily movement patterns (Peters et al., 2015). We estimated home ranges for broods with ≥30 radiotelemetry locations (Seaman & Powell, 1996). To do this, all environmental layers were clipped to an individual's buffered home range. Similar to a backward stepwise variable selection approach (Gherghel, Brischoux, & Papeş, 2018;Hastie, Tibshirani, & Friedman, 2001), we initially eliminated highly correlated variables (|r| ≥ 0.7; Dormann et al., 2013) and variables that had ≤5% contribution to model accuracy gain (Phillips, Anderson, & Schapire, 2006;Phillips & Dudík, 2008;Sahlean, Gherghel, Papeş, Strugariu, & Zamfirescu, 2014) for individual brood models. If two variables were highly correlated, we retained the variable that had the highest contribution to model accuracy gain. This resulted in idiosyncratic variable suites for each brood which were derived from our initial 22 environmental variables described above.
Individual Maxent models were initially run using default input values. This included the use of 10,000 background pseudo-absence points, a regularization multiplier of one, 500 iterations per model, and a convergence threshold of 0.00001. However, if an individual's buffered home range had <10,000 cells, we adjusted the number of background points to represent 90% of the cells within that home range. Models were replicated 100 times using a bootstrap method (Araújo, Marcondes-Machado, & Costa, 2014), and 25% of occurrence locations were held out as a test dataset to test the validity of our models (Sahlean et al., 2014). We used 10 percentile training presence as the threshold method to estimate the test omission error because this threshold method generally outperforms other threshold rules (Liu, White, & Newell, 2013).
Once all individual brood models were run, we used a Kruskal-Wallis test (Zar, 1999) to determine which variables were most influential in determining habitat suitability across all individuals by species (Baldwin, 2009). Variable influence was measured by each of their contributions to model accuracy gain, and the variables with the statistically greatest contributions to model accuracy gain (based on the results of the Kruskal-Wallis test) were retained. The variables that were retained after this step were then carried forward to build a first-order Maxent model to determine habitat suitability at the study site extent. All procedures described in this section were completed at both 2 m and 30 m grains.

| First-order Maxent modeling
A first-order Maxent model was created for both species at both 2 m and 30 m grains using the radiotelemetry occurrence data and the variables that were retained from the steps described in the previous section. All procedures described in the previous section were used to create first-order Maxent models. However, to account for model complexity, we calibrated our individual models by using different values for the regularization multiplier, which has been shown to significantly influence model performance (Radosavljevic & Anderson, 2014;Tanner, Papeş, et al., 2017). We compared average test omission errors across first-order models using regularization multiplier values of 0.25, 0.50, 1.00, 1.50, 2.00, 4.00, 6.00, 8.00, and 10.00 (Radosavljevic & Anderson, 2014). Models with different regularization multiplier values were repeated 10 times. We then tested for differences in the average test omission errors of these models through a one-way analysis of variance (ANOVA) and a post hoc Tukey HSD pairwise comparisons test (α = 0.05) using PROC GLM in SAS 9.4. The regularization multiplier value that resulted in the statistically lowest average test omission error was used for the final model for each species across both grains.
We used the logistic output from Maxent to obtain values of habitat suitability for each species across both grains. This ultimately created a map within the extent of our study site in which each cell had a probability of habitat suitability for each species. These values in theory could range from 0 to 1, with 1 representing a cell with 100% probability of habitat suitability for a specific species. These suitability values were then carried forward into our survival analyses to determine how habitat suitability influenced individual chick survival (described below).

| Habitat suitability analysis for nonbrooding adults
To determine whether brooding adults were differentially selecting habitat as a function of behavioral adaptations associated with brooding activities, we compared the habitat suitability indices created for brooding adults to habitat suitability indices created for nonbrooding adults using occurrence locations (i.e., radiotelemetry locations) of nonbrooding adults that were obtained during the brooding season for bobwhite and scaled quail (June 23-October 20 and June 9-October 12, 2013-2014, respectively). To create habitat suitability indices for nonbrooding adults, we used the Maxent algorithm and followed an identical protocol as described in the previous three sections. Thus, we obtained two unique nonbrooding habitat suitability indices for each species (i.e., 2 m and 30 m grains) and considered these nonbrooding indices to be representative of what was available to brooding adults based on what the rest of the population was using.
Once pairwise (brooding and nonbrooding) habitat suitability indices for each species at each grain were created, we estimated the amount of similarity between the indices to determine the measure of divergent space use associated with behavioral adaptations of parents with broods. To estimate similarity between indices, we estimated the relative rank (RR) metric (Warren & Seifert, 2011) through ENMTools (version 1.4.4;Warren, Glor, & Turelli, 2010). This metric is an estimate of the probability that any two patches of habitat have the same relative ranking for pairwise models (rather than quantifying similarity based on suitability values) and was used (as opposed to other similarity metrics for niche modeling) because we were interested in the relative prioritization of habitat across pairwise indices (Warren, Wright, Seifert, & Shaffer, 2014). Values range from 0 (low similarity of relative ranking) to 1 (complete similarity of relative ranking). We created a null distribution of the RR metric in ENMTools using the background similarity test function with 100 Maxent pseudoreplications that used a random sample of background pixels as occurrences instead of brooding and nonbrooding locations. We then determined whether the empirical RR values estimated from the pairwise brooding/nonbrooding index comparisons were contained within the null distribution of the RR metrics. If the empirical RR values were not within the null distribution, we considered the RR metric to be statistically significant (thus indicating differential space use between brooding and nonbrooding adults).

| Chick survival analysis
To estimate the daily survival probabilities of individual chicks by species, we used the nest survival model in Program MARK (version 9.0; Cooch & White, 2017;White & Burnham, 1999) as it allows for potential gaps in monitoring events and staggered entry of individuals into a population (Dinsmore, White, & Knopf, 2002). As the battery life of our transmitters was ~21-23 days, we estimated daily survival up to 20 days after transmitter attachment. Since chicks were caught ~8-12 days after hatch, this 20-day period after transmitter attachment was used to represent the stage of a chick's life history in which they are incapable of thermogenesis (Lusk, Guthery, Cox, DeMaso, & Peoples, 2005) and highly reliant on brooding adults for survival.
We attributed mortalities or signal losses that occurred within three days of transmitter attachment to be caused by transmitter failure or capture-related mortality. If this occurred, the associated chicks were censored from survival analysis (Larson, Clark, & Winterstein, 2001). When radio-tagged chicks were located at a distance >100 m from the brooding adult, we checked individual chicks for potential mortalities. If fates of chicks were unknown due to loss of transmitter signals, we right-censored individual history encounter files to the last date they were known to be alive.

Independence of sample units is an important assumption when
implementing survival analyses (Cooch & White, 2017) as a violation of this assumption could underestimate sample variances (Schmutz, Ward, Sedinger, & Rexstad, 1995). To test for independence of individual chicks within broods, we first used a bootstrapping simulation process in Program MARK to estimate an overdispersion factor (ĉ; Bishop, White, & Lukacs, 2008). A ĉ > 1 would be indicative of overdispersion and a lack of independence for individuals within groups, while a ĉ < 1 would indicate underdispersion with individuals being highly independent within groups (Bishop et al., 2008).
We initially built models that were representative of time trends.
This included the covariates linear time, quadratic time, and ordinal date of hatch. Global, additive, and univariate models were built using these three covariates, and the most parsimonious model was considered the model with the lowest Akaike's information criterion (AIC c ) value adjusted for small sample sizes (Burnham & Anderson, 2002). We then ran a bootstrapping simulation process on the most parsimonious model with 5,000 iterations (Chernick, 1999) and used a unique brood identification number to identify blocks of data. The ĉ values obtained from these bootstrap simulations were then applied to the remaining model-building strategies employed for the entire dataset.
Additionally, we included two variable suites representing daily weather conditions and landscape/vegetation composition (i.e., HSIs). Though our initial hypotheses were related to understanding how adaptive habitat selection of brooding adults influenced offspring survival, we tested effects of weather conditions because they have been shown to significantly influence habitat selection and survival of these species during brood rearing (Carroll, Davis, Elmore, Fuhlendorf, & Thacker, 2015;Terhune et al., 2019).
We obtained daily weather conditions from three weather stations (WeatherHawk 232) that were oriented west-to-east across our study site (mean distance between weather stations: 13 km). To do this, we used the weather station closest to the centroid of a brood's home range. Conditions were recorded every hour at each weather station. Weather variables included in our analysis were as follows: average daily ambient temperature (°C), maximum daily ambient temperature (°C), minimum daily ambient temperature (°C), average daily variance in ambient temperature, maximum daily solar radiation (W/m 2 ), average daily wind speed (km/hr), daily precipitation (mm), and average daily relative humidity (%). We conducted a principal component analysis (PCA) using singular value decomposition in Program R (v.3.4.3) to reduce the amount of variables and correlation among variables within our survival analysis. To determine the optimum number of principal components to retain in our analysis, we used the broken-stick stopping rule (Jackson, 1993). Based on this stopping rule, we retained two principal components (PCs) for our survival analysis. PC1 (44.7% of explained variance) represented hot dry days, while PC2 (18.8% of explained variance) represented cooler and more humid days with precipitation events (Appendix S2).
Finally, landscape/vegetation composition variables included six variables representing habitat suitability values derived from our final first-order Maxent models. To derive habitat suitability variables for our survival analysis, we used the logistic probability of suitability maps from our final first-order Maxent models and separated cells into categories using Jenks natural breaks classification (Jenks, 1967). We chose to limit the number of categories that habitat suitability values are grouped in by selecting the category in which the goodness of variance fit begins to increase at a decreasing rate (i.e., the point of inflection). This was to prevent an arbitrary selection for the number of categories we split our data into. For both species, across both grains, the point of inflection for goodness of variance fit values was three categories (Appendix S3). We arbitrarily refer to these categories as "low," "medium," and "high" habitat suitability values (Table 1). Once cells were classified, we calculated the percentage of each brood's home range that was composed of low, medium, and high habitat suitability cells and incorporated these values into our survival analysis.
We conducted a correlation analysis on all variables and eliminated any variables that were highly correlated (|r| ≥ 0.70) before our model-building exercise. The percentage of medium habitat suitability cells within a brood's home range were dropped for both species and both grains as these variables were highly correlated with other habitat suitability values. We employed a purposeful model-building strategy (Hosmer, David, Lemeshow, & Sturdivant, 2013) to identify the best performing model for chick survival by species, in which each species was modeled under a different framework. We initially assessed univariate models and retained variables with a p < 0.25 to build a global model. We then individually removed variables with a p > 0.05 from the global model based on the strength of the p value until a model contained only variables that had p < 0.05. Variables that were initially eliminated from our analysis were then added back to the reduced global model to determine whether the significance of the relationship had changed after incorporating other additive effects.
Once model building was complete, we determined the most parsimonious model based on the quasi-AIC c values (QAIC c ; Burnham & Anderson, 2002). We considered models with a ΔQAIC c < 2 to be competitive models for explaining variability in chick survival and considered variables that had β estimates whose 95% confidence intervals excluded 0 to be significant to chick survival (Burnham & Anderson, 2002). Daily survival rates were obtained from the most parsimonious model, and we used the Delta method (Powell, 2007) to compute 20-day survival rates and associated error rates.

| Post hoc parental survival analysis
We estimated daily survival probabilities of radio-collared bobwhite and scaled quail adults from June 23-October 20 and June 9-October 12, respectively. These two periods corresponded with the same temporal extent of the brood survival analyses associated with this study. To estimate survival probabilities, we use the nest survival model in Program MARK (White & Burnham, 1999). Our objective was to determine whether individuals that were involved with brooding activities were at greater risk of mortality than those that were not brooding. As such, we included three variables in the adult survival analysis: time, quadratic time, and the brooding status of the individual (i.e., either actively brooding or not brooding). Both male and female adults were included in our analysis as both sexes participate in brooding activities. If an individual ceased brooding activities, we right-censored their encounter history on the last day they were observed brooding and then created a new encounter history beginning on the next day they were observed without a brood (Mangelinckx et al., 2018). We implemented the same purposeful model-building strategy used in our chick survival analysis and compared explanatory models to a null model. Similarly, we considered models with a ΔAIC c < 2 to be competitive models for explaining variability in adult survival and considered variables that had β estimates whose 95% confidence intervals excluded 0 to be significant to adult survival (Burnham & Anderson, 2002). daily movement patterns were estimated. Though scaled quail broods had larger home ranges compared to bobwhite (scaled TA B L E 1 Low, medium, and high habitat suitability categories in which cells from rasters derived from population-level Maxent models were classified based on their individual values. Occurrence data for Maxent models represented radio-transmitter locations for northern bobwhite (Colinus virginianus) and scaled quail (Callipepla squamata) broods during the 2013-2014 breeding seasons at Beaver River WMA, Oklahoma, USA. Three categories were chosen based on Jenks natural breaks classification. Percentages represent the percent of our study site classified in the associated category

| Habitat suitability modeling for broods
There were no differences in the model performance of regularization multiplier values for both the 2 m (p = 0.14) and 30 m (p = 0.06) bobwhite brood first-order models. Therefore, a regularization multiplier value of one was used for both of these models. However, a regularization multiplier value of 10 was used for both the 2 m (p = 0.02) and 30 m (p = 0.02) scaled quail brood first-order models as these values were shown to have the best average test omission errors. Based on average test omission errors, first-order models performed well for both bobwhite (2 m = 0.11, 30 m = 0.14) and scaled quail (2 m = 0.10, 30 m = 0.10) broods.
Habitat suitability analyses at the first-order resulted in idiosyncratic variable suites for each brooding species across both grains (

| Chick survival
Our baseline temporal survival models suggested no support for temporal survival trends for bobwhite chicks (null model AIC c = 161.21, w = 0.28) and no support for temporal survival trends for scaled quail chicks (null model AIC c = 171.59, w = 0.39). Overdispersion factors calculated from these baseline temporal models suggested that there was variance inflation due to a lack of independence between individuals for scaled quail chicks (ĉ = 1.54), whereas underdispersion and independence were evident for bobwhite chicks (ĉ = 0.52).
The most parsimonious survival model for bobwhite chicks was The 20-day survival rate for bobwhite chicks was estimated as 0.91 (SE = 0.03) when a brood's home range was composed of 100% of high habitat suitability areas and decreased to 0.60 (SE = 0.11) when it was comprised of 0%.
For scaled quail chick survival, we did not detect any direct impacts of space use on chick survival as the null model was the best supported model (w = 0.20; Table 4). Though eight univariate models had ΔQAIC c < 2, all β coefficients for these parameters had confidence intervals overlapping 0. The overall DSR for scaled quail chicks was 0.9747 (SE = 0.0059), and the overall 20-day survival rate (0.5993, SE = 0.0720) for scaled quail chicks was statistically lower than that of bobwhite chicks. scaled quail as the null model was our best supported model ( Table 5).

| D ISCUSS I ON
However, the degree to which individuals modify their space use during this period can vary across space, time, and species. For instance, we found a greater degree of altered space use for brooding bobwhite when compared to brooding scaled quail (Figure 2).
Though life-history theory would suggest that both species should be expected to incur a cost due to these behavioral changes, we observed clear divergent rates of alternative reproductive strategies between species during the same time period , suggesting that interspecific differences in brooding strategies may be expected in sympatric species. Such relationships offer important insight into the conservation of these species during a vulnerable and important life-history stage.
Life-history theory suggests that there should be a trade-off between parental risk and offspring success for short-lived species (Ghalambor & Martin, 2001;Stearns, 1976). However, direct offspring success to such behavioral modifications from attending parents may be hard to quantify due to logistic constraints associated with marking individuals at such a vulnerable period, particularly for precocial species. By integrating both chick survival and adult survival simultaneously into our analyses, we attempted to eliminate any unobserved components of such a risk-to-reproduction tradeoff associated with these species. Such unobserved components have been posited as a potential source of ambiguity associated with studies assessing how adaptive habitat selection influences reproductive success and overall fitness (Bloom et al., 2013;Uboni et al., 2017 2002) or if a species is more likely to allocate resources for selfmaintenance. For instance, it has been suggested that scaled quail have adapted to arid and semi-arid regions (such as our study area) by allocating more resources toward self-maintenance rather than toward reproductive output whereas the opposite is true for bobwhite because they evolved under more mesic conditions (Giuliano, Patiño, & Lutz, 1998). Likewise, the null relationship between HSI and scaled quail chick survival could be driven by relationships in which environmental conditions influencing both species' occurrence and species' survival are disparate from the conditions only influencing a species' occurrence (Bacon et al., 2016;Ficetola, Miaud, Pompanon, & Taberlet, 2008). Furthermore, this null relationship could be related to the spatial scales that were used in our analyses (Bloom et al., 2013), though we attempted to account for this by using two grain sizes representing fine (2 m) and coarse (30 m) habitat configurations.
The scale at which organisms respond to environmental conditions and/or resources should vary across space and time if habitat selection is an adaptive process (Bloom et al., 2013;Wiens, 1989).
Furthermore, variation in the scale at which organisms select habitat will also change during different life-history stages (Addicott et al., 1987;Chalfoun & Martin, 2007;McGarigal et al., 2016) which could potentially decouple the link between space use and  [Johnson, 1980]) for an individual is constrained during a specific life-history stage such as brood rearing (Carroll et al., 2015) due to decreased movement abilities or physiological constraints, a brood-rearing adult may not be able to select certain areas on a landscape that increases brood survival.
We found that the brooding habitat selection-fitness relationship is dynamic across spatial grains, as the relationship between bobwhite chick survival and HSI was only significant at the 30 m grain. In this study, we used an index of habitat suitability derived conditions considered in a model and is commonly used as a tool for conservation purposes (Guisan & Thuiller, 2005). However, the relationship between habitat suitability and demographic parameters related to population persistence is often complicated and disparate across species, space, and time (Bacon et al., 2016).
Specifically, our results suggest that the relationship between habitat suitability and chick survival can vary across species and spatial scale. Therefore, if such indices are to be used as a tool for conservation efforts, single-scale suitability indices may be too simple and miss such complex relationships (Bacon et al., 2016).
Our research illustrates the importance in understanding the complex relationships between parental space use, habitat suitability, and how these interact to influence chick and attending parental survival. It is evident that these relationships exist in a complex state that can be species-specific, spatially variable, and potentially influenced by life-history strategies. Indeed, for bobwhite in our study, there was a clear cost of parental attendance in exchange for offspring survival, which can help explain other aspects of their life-history strategies. For example, brood amalgamation on our study site was more common in bobwhite broods than scaled quail broods at the chick life stage (Orange, 2015). Such breeding behavioral strategies have been postulated to offer survival benefits for the offspring (Dahlgren, Messmer, & Koons, 2010;Eadie, Kehoe, & Nudds, 1988;Lott & Mastrup, 1999), yet could have implications into understanding a link between demographics and differential habitat selection. Furthermore, when assessing demographic parameters such as precocial chick survival, these relationships are hierarchical in that it is ultimately tied to the brooding adult's behavioral adaptations. Evidence emphasizes the importance of chick survival for population persistence of r-selected species (Burger, Dailey, Kurzejeski, & Ryan, 1995;Colwell et al., 2007;Ludwig, Aebischer, Bubb, Roos, & Baines, 2018;Sandercock et al., 2008;Terhune, Sission, Grand, & Stribling, 2007), and if concepts such as habitat selection and habitat suitability are to be used for conservation purposes (Morrison et al., 2012), efforts to understand simultaneous trade-offs between offspring survival, adaptive habitat selection, and parental risk must be considered.

ACK N OWLED G M ENTS
We thank our field technicians for their logistical support while conducting research in the field. We also thank M. Carroll

CO N FLI C T O F I NTE R E S T
The authors declare no conflict of interest.

AUTH O R CO NTR I B UTI O N S
All authors contributed equally to this manuscript.

DATA AVA I L A B I L I T Y S TAT E M E N T
Data available from the Dryad Digital Repository: https ://doi. org/10.5061/dryad.2ct6558.