2.1 Study area

Our study focused on the Bi-State DPS (16,880 km²), which spans portions of the border of Nevada and California and is situated in the topographically diverse area between the Sierra Nevada and the Great Basin. Vegetation predominantly consisted of mountain big sagebrush (Artemisia tridentata vaseyana), with some areas of low sagebrush (A. arbuscula), Wyoming big sagebrush (A. t. wyomingensis), silver sagebrush (A. cana), and basin big sagebrush (A. t. tridentata) occurring locally. Invasive cheatgrass (Bromus tectorum) was present but uncommon. Two native conifers, single-leaf pinyon (Pinus monophyla) and Utah juniper (Juniperus osteosperma), occurred at higher elevations and in mesic sagebrush environments.

2.2 Field methods

Adult and juvenile sage-grouse were captured across 10 subpopulations within the Bi-State DPS (Appendix S1) from 2003 to 2019 in the spring, summer, and fall seasons using standard spotlighting techniques and were classified by age and sex using published methodologies (Ammann, 1944). Captured sage-grouse were fitted with either necklace-style very high-frequency radio-transmitters (Advanced Telemetry Systems, Isanti, Minnesota) or rump-mounted Global Positioning System platform transmitting terminals (GeoTrack, Apex, North Carolina). All sage-grouse were captured and handled in accordance with the USGS Western Ecological Research Center (WERC) Animal Care and Use Protocol WERC-2015-02.

Radio-marked birds were located using portable receivers and handheld Yagi antennas during nesting and brood-rearing periods ≥2 times per week. To identify nest locations, we avoided flushing females and visually confirmed nesting status for females that were relocated in the same locations during two consecutive checks, which indicated that incubation had begun. Nests were considered successful if ≥1 chick hatched, as determined by a visual assessment of eggshell remains or by observing chicks in the nest bowl. Females with successful nests were then located and checked for brood presence every 10 days for up to 50 days, with checks conducted both during daylight and nighttime hours, which allowed the presence of chicks with the marked female to be checked with spotlights. At 50 days posthatch, after which young are capable of surviving independently, the brood was flushed and considered successful if ≥1 chick survived.

2.3 Spatial data

We evaluated the effects of multiple spatially and temporally explicit environmental covariates on sage-grouse habitat selection and survival across reproductive life stages (Table 1). For temporally varying covariates, we used data from the year a nest or brood was monitored. We evaluated the selection of land cover variables at multiple spatial scales because animals often select habitat in a scale-dependent manner (Boyce, 2006). For all analyses, we calculated the proportion of each land cover covariate or density of feature (e.g., streams) within circular moving windows with radii of 167.9, 439.5, and 1451.7 m, which represented averages of minimum, mean, and maximum daily distances traveled by sage-grouse (Coates, Casazza, et al., 2016). We included additional scales for nest and brood analyses based on the assumption that movement patterns would differ during these life stages because a female is tied to either a single nest site or a brood with limited mobility. We included a finer scale (radius = 75 m), representing the core recess area used by nesting females, and an intermediate scale (260 m), representing the average distance moved during nesting recesses (Dudko et al., 2018). We included an additional scale that represented the median movement distance of a female with broods within either the early or late brood-rearing period (early: 260 m; late: 370 m, Coates et al., 2024). For distance-based predictors, we transformed distances using an exponential decay function, where $$$ e=-d/\alpha $$$ and $$$ \alpha $$$ was the mean value at all locations for a given analysis (Coates, Casazza, et al., 2016). This allowed for the effect of a predictor to decay with increasing distance. All variables were scaled and centered prior to analysis.

2.4 Analysis

Relationships between wildlife and habitat are typically scale-dependent, and rarely does a single “best” scale for all habitat variables exist (Stuber & Fontaine, 2019). Therefore, we first used Bayesian latent indicator variable scale selection to select the most influential scale among each group of variables (BLISS; Stuber et al., 2017). BLISS uses latent scale indicator variables estimated with reversible-jump Markov chain Monte Carlo (MCMC) sampling to select the best scale within a group without issues of collinearity (Stuber et al., 2017). Variables were grouped into 12 categories to represent similar features and correlated variables (|r| ≥ .5), such that no variables between groups were highly correlated (Table 1).

We then developed a final model with only the most influential variables from each group to allow for straightforward inferences. We calculated the probability of direction (p(|β| > 0); Makowski et al., 2019) and considered probabilities ≥.85 and ≥.95 to represent moderate and strong effects, respectively. For each variable in the final selection models, we calculated relative selection strengths (RSS), which provide an estimate of the relative intensity of two locations differing by one standard deviation, assuming that both locations were equally available (Avgar et al., 2017). An RSS >1 represented a positive effect on habitat selection, whereas an RSS <1 represented a negative effect.

We fit all models using MCMC simulations with JAGS (Plummer, 2003) implemented via the “R2Jags” package (Su & Yajima, 2015) using Program R (R Core Team, 2021). Vague normal priors were used for random effects and their measures of error (μ = 0, σ² = Uniform (0, 100); Kéry, 2010). To prevent overfitting, we specified Lasso (i.e., Laplace) prior distributions for each habitat predictor, with an uninformative hyperprior specified for the tuning parameter lambda (Uniform(0.001, 10); Park & Casella, 2008; Hooten & Hobbs, 2015). Each model ran for 30,000 iterations, thinned by a factor of 5, and we discarded the first 20,000 samples and made inferences based on the remaining 10,000 samples from 3 independent MCMC chains. We assessed convergence and MCMC chain mixing visually and based on Gelman-Rubin convergence statistics (<1.1; Gelman & Hill, 2006).

2.5 Model validation

After choosing a final model for each reproductive life stage, we used independent testing data from 50 individual birds or broods randomly withheld to validate the model's ability to predict nest and brood locations. We used cross-validation methods for RSFs (Johnson et al., 2006) and used-habitat calibration (UHC) plots (Fieberg et al., 2018) to validate selection models. UHC plots compare observed distributions for each habitat variable to the expected distribution based on the model (Fieberg et al., 2018), using the fitted RSF to generate a predictive distribution of used habitat, which is then compared to the independent testing data for each habitat variable. For survival models, we generated survival predictions based on the posterior distributions from the final model to create predicted survival curves which were then compared to the observed survival curves from the independent testing data, which resulted in a post-hoc Bayesian predictive p-value (Gelman et al., 2013), where values near 0 or 1 indicate poor model fit (Schmidt et al., 2010).

2.6 Selection and survival mapping

Using estimates from the final models and the spatial data layers described above, we mapped both selection and survival for each reproductive life stage across the Bi-State DPS (Coates et al., 2024). We assumed that the median value of the posterior distribution represented the best estimate for each covariate and applied these in the model equations outlined above, where the matrix X represented the relevant raster values for each covariate across each 900 m² pixel in the Bi-State DPS. For time-varying covariates, we used the raster values from 2021, which represented the most recent layers available. We did not include distance to lek or nest in the habitat selection maps because we were primarily interested in the spatial distribution of areas suitable for sage-grouse based on underlying habitat characteristics. Including distance to lek would provide information on the distribution of occupied habitat, which we addressed in a map of example management categories (described below) following similar methods to Coates, Casazza, et al. (2016).

For selection maps, we transformed estimates using a habitat selection index (HSI), where $$$ \mathrm{HSI}=\frac{w(x)}{1+w(x)} $$$, which indicates relative habitat use proportional to availability on a scale of 0–1 (Coates, Casazza, et al., 2016). We then categorized the continuous selection surface into four categories (nonhabitat, low, moderate, and high selection) using the percent isopleth method at used locations with cutoff values at the 50th, 25th, and 5th percentiles (Doherty et al., 2016; O'Neil et al., 2020). To make population-level inferences on survival, we exponentiated values of daily survival based on the number of days in either the nesting or brood-rearing periods (nesting = 38 days, early brood-rearing = 21 days, late brood rearing = 28 days) to calculate cumulative survival for each period. We classified the continuous survival surface for each life stage into four categories (very low, low, moderate, and high survival) based on the distribution of values at failed and successful points following O'Neil et al. (2020).

We then used continuous selection and survival maps for each reproductive life stage to calculate composite selection and survival indices representative of the entire reproductive period. To calculate the composite selection index, we first relativized the habitat selection indices described above by dividing by the maximum value in that life stage (Coates, Brussee, et al., 2020), weighting all three reproductive life stages equally when calculating a composite surface. We then multiplied the three relativized surfaces together to create a composite selection index, which we categorized using the method described above. Calculating the survival index followed a similar procedure, but values were not relativized because the outputs represented true survival probabilities. We multiplied the exponentiated survival surfaces that represented overall survival during a given life stage to create a composite survival index, which we classified following the same methods used for individual survival maps. We only calculated the survival index for pixels that were considered high, moderate, or low based on the composite selection index, thus excluding areas considered nonhabitat.

We then created a ranked index that represented the overlap between the categorized selection and survival maps for each life stage and the composite reproductive indices. The highest habitat rank occurred in pixels that had both high selection and survival and thus were assumed to represent productive habitat for that life stage. In contrast, the lowest rank occurred in pixels with high selection but very low predicted survival, which could indicate maladaptive habitat selection.

Finally, we mapped specific example management categories representing the overlap between the composite selection index and an abundance and space use index (ASUI) calculated from breeding surveys (Coates, Casazza, et al., 2016), which allowed us to highlight high-quality habitat currently occupied by sage-grouse. The abundance and space use index represented the distribution of habitat predicted to be occupied by sage-grouse by accounting for the configuration of leks, the distance to lek, and the predicted abundance at each lek over time. By combining the abundance and space use index with the selection and survival maps described here, we were able to differentiate between likely occupied habitat and potential habitat with low or no occupancy. We defined four example management categories (Coates, Casazza, et al., 2016): (1) priority+ areas, defined as important source habitat for each reproductive life stage supporting high selection and high survival within high-use areas based on the ASUI, (2) priority areas, defined as areas with either predicted high, moderate, or low selection (excluding nonhabitat) and high use based on the ASUI or source areas outside of high use areas based on the ASUI, (3) general areas, defined as the intersection with high selection and low to no use based on the ASUI or nonhabitat within areas of high use based on the ASUI, and (4) other areas, defined as moderate selection within areas of low to no use based on the ASUI. We provide data to support all models, along with resulting raster surfaces to support land management and conservation efforts, via the USGS ScienceBase digital repository (Coates et al., 2024).

3.1 Nest selection and survival

For nest sites, the relative probability of selection increased with greater sagebrush height and at higher elevations (Figure 1, Appendix S2). The relative probability of nest site selection decreased in areas with more herbaceous cover, pinyon-juniper cover, bare ground, burned area, streams, springs, saline lakes, and south-southwesterly aspects (Figure 1). Our nest selection model had reasonable out-of-sample predictive capabilities (Spearman's rank ρ = .98, R² = .85, β_predict = .64) and used-habitat calibration plots using an independent dataset demonstrated that model-predicted habitat was consistent with observed nest locations (Appendix S2).

Nest survival was higher with more burned area, more shrubs, more topographic roughness, and at higher elevations (Figure 2, Appendix S2). Selection and survival were misaligned for burned areas. However, data from within burned areas was limited, with only 2% of nests within a fire perimeter, so we interpreted the moderately positive relationship with caution. The predicted survival curves from the model were consistent with the testing data (Bayesian p-value = .55).

3.2 Early brood selection and survival

The relative probability of selection during early brood rearing increased with greater sagebrush height, more bare ground, and at higher elevations (Figure 1, Appendix S2). The relative probability of selection during early brood-rearing decreased with more wet meadows, pinyon-juniper cover, more cumulative burned area, more perennial streams, greater topographic roughness, and a higher heat load index (Figure 1). Our early brood selection model had strong out-of-sample predictive capabilities (Spearman's rank ρ = .99, R² = .96, β_predict = 1.00), and used-habitat calibration plots using an independent dataset demonstrated that model-predicted habitat was consistent with observed early brood locations (Appendix S2).

Early brood survival increased with more burned area and closer to conifer cover (Figure 2, Appendix S2), and decreased with more shrubs and more perennial streams (Figure 2). The predicted survival curves from the model were consistent with the testing data (Bayesian p-value = .55). Selection and survival differed with moderate or strong effects for shrubs, burned area, and conifer cover. Broods selected for areas with taller sagebrush, but survival was higher in areas with lower shrub density. Early broods were more likely to select areas with less burned area closer to conifer cover, which was associated with lower survival, although there was high uncertainty around the effect of conifer cover on early brood survival.

3.3 Late brood selection and survival

The relative probability of selection during late brood rearing increased with greater sagebrush height, more herbaceous cover, and at higher elevations (Figure 1, Appendix S2). The relative probability of selection during late brood rearing decreased with greater pinyon-juniper cover, more bare ground, more cumulative burned area, more perennial streams, more saline lakes, steeper slopes, and more south-southwesterly aspects (Figure 1). Our late brood selection model had reasonable out-of-sample predictive capabilities (Spearman's rank ρ = .95, R² = .62, β_predict = .60), and used-habitat calibration plots using an independent dataset demonstrated that model-predicted habitat was consistent with observed late brood locations (Appendix S2).

Late brood survival increased with more intermittent streams, more springs, steeper slopes, and higher values of the compound topographic index (Figure 2, Appendix S2). The predicted survival curves from the model were consistent with the testing data (Bayesian p-value = .55).

Two variables differed across selection and survival with moderate or strong effects: streams and slope. Selection was lower but survival was higher in places with more perennial or intermittent streams, respectively. During late brood-rearing, selection was higher but survival was lower in areas with steeper slopes.

3.4 Mapping

We mapped selection (Figure 3) and survival (Figure 4) across the Bi-State DPS. We calculated composite selection and survival indices across the Bi-State DPS by combining the selection and survival maps for the 3 reproductive life stages. Based on the composite selection index, a total of 5.3% of the Bi-State region was classified as high selection (844 km²), 15.2% as moderate (2419 km²), 41.2% as low (6536 km²), and 38.3% as nonhabitat (6325 km²; Figure 3). Of the area classified as habitat, 17.2% of the Bi-State region was classified as high survival (1684 km²), 19.9% as moderate (1947 km²), 14.8% as low (1445 km²), and 48.1% as very low (4621 km²; Figure 4). We also created combined selection-survival maps across the Bi-State DPS by combining categorized selection and survival maps. Productive habitat (high or moderate selection paired with high or moderate survival) represented 14.1% (1375 km²), whereas areas with potentially maladaptive selection (high or moderate selection paired with low or extremely low survival) represented 19.3% of the Bi-State DPS (1888 km²; Figure 5). We also mapped specific example management categories using the composite selection map and the abundance and space use index (see Appendix S2).

4.1 Wildfire and annual grass

Wildfire is currently reshaping ecosystems across western North America (Flannigan et al., 2009), and our results suggest mismatches in selection and survival consequences for burned areas. During brood rearing, sage-grouse avoided burned areas despite experiencing higher survival in those areas. However, only 2% of nests were within fire perimeters, and the apparent avoidance during the brood-rearing period may be partially due to a lack of fires in areas accessible to sage-grouse broods. While previous studies have found negative effects of wildfire on sage-grouse across multiple life stages (Anthony et al., 2022; O'Neil et al., 2020), this has largely been attributed to recent megafires (Anthony et al., 2022; Dudley et al., 2021) coupled with the feedback loop between fire and invasive annual grasses, which has had the strongest negative effects in other parts of the Great Basin (Brussee et al., 2022; Coates, Ricca, et al., 2016; Germino et al., 2016). In comparison, megafires are rare in the Bi-State, and invasion by annual grasses is uncommon (only 4.5% of burned areas had >10% annual grass cover). In addition, previous studies have found positive short-term effects of wildfire on brood survival when cheatgrass is not present, likely due to increased herbaceous cover and forb production (Brussee et al., 2022), which is consistent with our findings. Given the large-scale threat posed by wildfire in sagebrush ecosystems, identifying productive habitats that support high selection and high survival across multiple life stages, especially with respect to burned areas, will likely be key to guiding habitat management under changing future conditions.

4.2 Conifer cover

Increased conifer cover has negative consequences for the population performance of sage-grouse (Olsen et al., 2021), and our results indicated strong avoidance of even low amounts of pinyon-juniper cover across reproductive life stages. During nesting, sage-grouse strongly avoided low levels of pinyon-juniper encroachment, but the strength of the avoidance decreased during early and late brood-rearing. Adult survival was lower in areas with more conifer cover (Coates et al., 2017), while conifer removal improved brood success (Sandford et al., 2017). Particularly at low levels of encroachment when shrubs are still dominant, conifer cover can represent an ecological trap where resources appear attractive but survival is reduced, in part due to the increased presence of predators such as perching raptors (Coates et al., 2017). Due to strong avoidance across all life stages, however, we found little to no evidence of an effect of conifer cover on survival, although there was high uncertainty around the effect on early brood survival. Reduced avoidance later in the season is likely due to trade-offs where broods must move up in elevation to access limited mesic areas that provide important cover and forage resources (Saher et al., 2022). However, high-elevation areas also tend to have greater pinyon-juniper cover, thus making it more difficult for broods to avoid conifers, potentially with negative survival consequences, although our results did not support impacts on survival. Overall, this suggests that sage-grouse may face important trade-offs during reproductive life stages, wherein avoidance of risk from predators may be prioritized during nesting and early brood rearing. Access to forage resources, such as those in high-elevation mesic areas, is likely most important during the late brood-rearing period when chicks are larger, risk is less of a factor, and resources become increasingly limited. If such trade-offs are highlighted and accounted for, our results suggest targeted restoration activities such as conifer removal in places like high-elevation mesic areas that birds are selecting for but where they are experiencing reduced survival would be highly beneficial for sage-grouse.

Our study was not without limitations. Perhaps most importantly, we did not incorporate movement dynamics because we only looked at how individual pixels were ranked across different life stages. Nevertheless, individuals that do not have to move large distances to meet their life stage habitat requirements are likely to be more successful, and our analysis identifies areas that benefit multiple reproductive life stages, allowing us to highlight both mismatches between selection and survival and trade-offs between life stages. In addition, we modeled brood survival, which is a binary presence/absence data type, rather than chick survival, which inevitably leads to a loss of information that could confound our results. When possible, modeling chick survival would be preferred, although marking and tracking individual chicks comes with additional challenges that we could not overcome in this study. We also lacked sufficient data on some features to reliably estimate effects, in large part due to a lack of those features on the landscape within our study area, but that inevitably limited our inferences regarding specific features. For example, only 2% of nests were within fire perimeters, and no nests or broods were located near agricultural cropland. In addition, some temporally varying covariates, such as precipitation, invariably affect habitat quality but are difficult to adequately capture in a habitat selection analysis. However, evaluating both habitat selection and survival across many years, as in our study, should provide useful relative indices across time. Finally, by limiting our study to the breeding season, we were only able to identify areas that may represent sinks for a single life stage rather than population-level sinks. Nonetheless, reproduction has a large impact on sage-grouse population dynamics (Taylor et al., 2012), and we would expect areas supporting lower survival during the nesting and brood-rearing periods to have correspondingly lower population growth rates.