Galliformes are challenging to keep alive and breed in captivity (Lance et al., 1970; Flieg, 1971; Johnsguard, 1983; Hancock, 1993), and their post-release survival and integration into wild populations has proved to be even more difficult (Parker et al., 2012; Carrlson et al., 2014; Mathews et al., 2016). Translocations of captive-bred Galliformes to wild environments generally result in few individuals surviving to reproduce the subsequent year (Sokos et al., 2008; Wiechman et al., 2011; Collar, 2020). Most die within a few weeks post release, primarily from predation (Parish and Sotherton, 2007; Merta et al., 2016), partly as a result of poor predator avoidance behaviors of inexperienced birds (Rantanen et al., 2010). High post-release mortality may also be due to suboptimal habitat selection and lower fitness (Rantanen et al., 2010), physiological stress during both pre- and post-release periods (Dickens et al., 2009), and exposure to weather, malnutrition and (or) starvation (Thompson et al., 2015). When captive-bred Galliformes do survive to reproduce, they can produce fewer offspring relative to their wild counterparts (e.g., 0.05 hatched chicks per released female vs. 2.09 hatched chicks per wild female, Putaala and Hissa, 1998) due to hen and egg predation (Putaala and Hissa, 1998; Parish and Sotherton, 2007) and suboptimal habitat selection and higher nest abandonment (Sage et al., 2003).

Galliformes, especially Greater sage-grouse; hereafter, sage-grouse), are known to reside in suboptimal locations where they may not perceive risk and modify their movements to optimize their fitness (Aldridge and Boyce, 2007; Heinrichs et al., 2018; Pratt and Beck, 2021). Hence, knowledge of how sage-grouse navigate through the risk landscape could reveal mismatches in perceived and actual risk (Pratt and Beck, 2021). This information may be useful in avoiding the assumption that translocated sage-grouse perceive risk as mapped or confirming they are responding to risk in a novel wild environment (Parker et al., 2012). This information would also inform and direct recovery planning and decision-making regarding habitat management that would optimize resources and maximize conservation gain (i.e., focus improvements where most valuable).

Like other Galliformes, sage-grouse were historically abundant throughout North America, but declined precipitously during the 20^th century due to habitat conversion, degradation, fragmentation, and reduction of sagebrush (Artemisia spp.) (Knick et al., 2003; Connelly et al., 2004; Aldridge et al., 2008). In Canada, only two small isolated populations remain in Alberta and Saskatchewan, and sage-grouse have declined by up to 92% in the last ~30 years (Environment Canada, 2014). As such, sage-grouse are considered endangered in Canada under the Canadian SARA (Environment Canada, 2014), prompting an urgent response including a captive breeding and release program to augment wild populations. Adaptive improvements to both the program and release environment are ongoing to support the recovery of sage-grouse in Canada. Habitat restoration, reclamation, infrastructure removal, beneficial grazing, and a multitude of other practices have been identified in recovery plans and implemented on the ground (AESRD, 2013; Parks Canada Agency, 2021).

Sage-grouse are sagebrush-obligate species and use sagebrush throughout much of their life cycle for food and cover. Although the species once flourished in Canada, human-induced changes to the environment (e.g., vegetation, predator guild, disturbance) have contributed to sage-grouse population declines. Sage-grouse currently exist in a mosaic of public and private lands in Canada, and their survival across the species' range is affected by a range of environmental factors and stressors including anthropogenic features (e.g., energy infrastructure, roads, buildings; Aldridge and Boyce, 2007), agricultural development and activities (Beck et al., 2006), predators (Tack, 2009), and insufficient habitat (e.g., inadequate sagebrush, shrub and grass cover; Connelly et al., 2004; Gregg and Crawford, 2010; rugged terrain, Beers and Frey, 2022). Large portions of the species’ habitat and current range are protected, and critical habitat has been identified and mapped using a resource selection function (Environment Canada, 2014). In Canada, sage-grouse critical habitat includes areas with large patches of moderate and spatially-dispersed shrub cover (e.g., silver sagebrush), minimal bare ground, moderately moist habitats with some amount of lush green vegetative cover, adequate availability of forage forbs, as well as areas with lower intensity of anthropogenic disturbance (e.g., human settlements, roads, fences, annual cropland, artificial structures for avian predators (Environment Canada, 2014).

We assessed mortality risk using animal location and known fate data from captive-bred sage-grouse that were released at two release sites (near Manyberries, Alberta; and west block of Grasslands National Park near Val Marie, Saskatchewan; Figure 1 ). Both study sites are within the mixed-grass prairie ecoregion (Alberta Parks, 2015; Parks Canada, 2022), where silver sagebrush (A. cana) is the dominant shrub species (see Supplementary Materials for further details).

In Alberta and Saskatchewan, sage-grouse habitat includes several potential stressors or risks including avian predator perches (e.g., trees, powerlines, fence posts), habitat loss (e.g., lack of shrub cover), and environmental disturbance (e.g., energy infrastructure, increased vehicle traffic) (Environment Canada, 2014). Cattle grazing occurs across much of the sage-grouse range on private lands, provincially designated grazing leases, and federally designated grazing areas. The Alberta and Saskatchewan study areas differ in the composition and spatial distribution of risk factors. For instance, much of the sage-grouse habitat in Alberta has been affected by oil and gas extraction activities, whereas anthropogenic infrastructure is relatively limited in Grasslands National Park in Saskatchewan (Government of Saskatchewan, 2023). In Grasslands National Park, habitat is concentrated in the valley bottom, and limited in the West block of the park where sage-grouse were released (Thorpe and Stephens, 2017), with escarpments separating lowland and flat upland habitats. By contrast, sage-grouse habitat occurs across a larger area with varied topography in Alberta.

Juvenile sage-grouse were released in Canada from 2018 to 2022 (n = 228), in Alberta (n = 119) and in Saskatchewan (n = 109). Sage-grouse were weighed, sexed, and tagged with transmitters before release and tracked until death or signal loss. Many sage-grouse mortalities occurred soon after release and proximate to the release site. These individuals had little opportunity to move away from the release site and select landscape features in the risk landscape. As our aim was to spatially understand the environmental covariates contributing to longer survival, we restricted analyses to individuals that survived at least 10 days after release. This also minimized the noise associated with the post-release ‘settling down’ period (Musil et al., 1993).

We categorized mortality as early (e.g., survived ≤ 10 days post release) and late (e.g., survived > 10 days post release), based on median survival days. We used the median survival as a threshold for including individuals (and all associated movement records) in the analysis, and exclusion of individuals that died too early to experience much of the post-release landscape. We used the median rather than average survival, because most sage-grouse died shortly after release, resulting in a right-skewed distribution. When carcasses were found, a cause of death was determined with either high or low confidence by biologists at the Wilder Institute at the Calgary Zoo ( Supplementary Table 1 ). We analyzed bird location data separately for each province of release and built models with both spring and fall release seasons combined, and only fall releases (i.e., in September and October).

To assess survival risk relative to environmental features, we developed a range of spatial data layers to use as covariates in our models. We selected similar types of covariates for each study site ( Supplementary Tables 2–4 ) and developed layers to describe each family of covariate (e.g., anthropogenic disturbance, cover of shrubs and sagebrush, critical habitat, release site, roads, terrain, vegetation, vertical features, water). We included a map of critical habitat (Environment Canada, 2014), derived from a sage-grouse nest occurrence model for Alberta (Aldridge and Boyce, 2007), that was extrapolated to Saskatchewan. The resource selection function included covariates describing mean shrub cover within 1 km², proportion of patchily distributed shrub cover within 1 km², greenness and wetness indices, and relationships to roads and other built-features.

For each spatial variable, we derived proportion, proximity, or cover layers, summarized at various spatial scales to evaluate the scale of influence of each variables. For example, we summarized proportional data using two moving window sizes that are biologically relevant to sage-grouse. In Alberta, we tested survival responses to proportional covariates using windows sizes of 0.55 km, the median daily distance traveled by birds released in Alberta between 2018 and 2022, and a 1-km survival-infrastructure responses from previous research (e.g., Aldridge and Boyce, 2007). In Saskatchewan, we similarly evaluated the median daily distance traveled by birds released in Saskatchewan (2.3 km), and tested responses using a 0.56 km window used in other sage-grouse studies (e.g., Aldridge and Boyce, 2007; Fedy et al., 2014; Kirol et al., 2015; Parsons et al., 2022). We evaluated the predictive ability of each scale in univariate mortality risk analyses and retained the scale of the covariate that had the lowest Akaike Information Criteria corrected for small sample size (AICc, Burnham and Anderson, 2002).

Our primary focus was assessing spatial features influencing sage-grouse mortality risk; however, species traits (e.g., sex) and release features (e.g., timing, age at release) can also influence mortality (Swenson, 1986; Beck et al., 2006; Maness and Anderson, 2013). We included the following aspatial covariates in statistical analyses: acclimation period (e.g., > or < 14 days), age at release, body mass, raising method (e.g., hen- and hand-raised), release period (e.g., early release before October 17, and late release after October 17), release type (e.g., hard, modified soft, and soft), and sex (see Supplementary Materials for hypotheses).

We implemented a four-step approach to assess sage-grouse movements and mortality risks associated with environmental covariates ( Figure 2 ). In step one, we conducted a univariate screening between covariates and mortality risk, to reduce the number of covariates entering multivariate models to accommodate sample size constraints. Prior to fitting models, we assessed correlation among covariates (Pearson’s r ≥ |0.70|). We standardized covariates by subtracting the covariate mean from the individual value and dividing it by the standard deviation. In step two, we built multivariate models of remaining covariates to assess time to mortality. In step three, we selected the top performing models and applied the multivariate equation to spatially referenced covariates to create a risk surface map. In step four, we fit a step-selection function using the sage-grouse GPS location data, risk surfaces (from step three), and existing critical habitat maps.

Female sage-grouse generally had a higher number of confirmed mortalities than males, likely due to females being easier to track (e.g., had larger, more powerful transmitters than males), but typically lived longer than males ( Supplementary Tables 5, 6 ). Time to mortality for females varied between provinces. With few exceptions, sage-grouse died within 20 days of release in Alberta and within 26 days in Saskatchewan ( Figures 3A, C ). Only 30% of males and ~58% of females survived past 10 days after release in Alberta; and a greater proportion of the released population survived past 10 days in Saskatchewan (50% of males, ~78% of females; Supplementary Tables 5 , 6 , respectively). Only two hens survived more than one year in our study sites (e.g., Alberta hens = 873, and 1,185 days – dropped transmitter). These outliers were omitted from analyses, but we found both hens primarily used less risky habitat (mean risk = 0.10, SD = 0.07, maximum risk value = 0.36, n = 152 locations; mean risk = 0.13, SD = 0.06, maximum risk value = 0.34, n = 207 locations; respectively) compared to other sage-grouse that survived at least 10 days (mean risk = 0.23; SD = 0.15; maximum risk value = 0.75; n = 1,220 locations). Most individuals had known fates ( Supplementary Tables 5, 6 ). Unknown survival status resulted from transmitter failure or dropped devices. Location data were used in analyses until the time the signal was lost (see Multivariate Survival Model methods above).

Predation accounted for most confirmed deaths in both study sites ( Figures 3 , 4 ; Supplementary Table 2 ). Carcasses were examined by biologists and the predators associated with mortalities were identified as mammal (e.g., canids such as coyotes, Canis latrans) and avian species (e.g., raptors such as great horned owls, Bubo virginianus) in both study sites ( Figure 4 ). Confirmed mortalities were also caused by weather (exposure) in both provinces. There was a notable snowstorm in Alberta in November 2020, which resulted in the accumulation of ~45 cm of snow. The extreme, extended temperatures (e.g., -20°C on ~90% of nights during a 2-week period in December 2021 and January 2022) and predation contributed to sage-grouse deaths in Alberta. Mortalities were also caused by collisions with vehicles and fences in Alberta only ( Figure 4 ; Supplementary Table 2 ).

We fit 83 univariate Cox proportional hazards and Accelerated Failure Time models for Alberta ( Supplementary Table 9 ), which resulted in nine covariates that moved into multivariate survival models (see Table 1 for subset of candidate models). Of those nine covariates, distance to gravel roads and streams strongly influenced mortality in Alberta ( Table 1 and Supplementary Table 9 ). For Saskatchewan, we fit 61 univariate Cox proportional hazards and Accelerated Failure Time models ( Supplementary Table 13 ), which resulted in 10 covariates that moved into multivariate survival models (see Table 1 for candidate models only). Of those 10 covariates, anthropogenic disturbance density within 0.56 km, proportion of meadow and other shrubs within a median daily distance travelled (2.35 km), elevation, and distance to roads and tall features strongly influenced mortality in Saskatchewan ( Table 1 and Supplementary Table 13 ). In univariate tests, acclimation period and raising method had weak predictive ability and were not retained in multivariate survival models ( Supplementary Tables 7–19 ; Figures 1 , 2 ). Later release dates (e.g., after October 17), influenced mortality risk, but only in Saskatchewan ( Supplementary Table 13 ).

We initially started with a list of spatial and aspatial covariates in our multivariate survival models, but spatial factors outcompeted non-spatial covariates ( Supplementary Tables 7–19 ; Figures 1 , 2 ). In the Alberta multivariate survival model, we included 46 sage-grouse (10 males and 36 females, n = 878 total locations) ( Table 1 ). The top multivariate survival model included distance to anthropogenic disturbance, which strongly influenced sage-grouse mortality risk ( Table 1 ). We found mortality risk strongly decreased by 62% per 1-km increase in distance to anthropogenic disturbance (risk ratio: 1.00 – exp(β) = 0.38, SE = 0.44, 95% CI = 0.17, 0.88, Table 2 ). The proportion of critical habitat (within a median daily distance travelled; 0.55 km) had a weak effect ( Tables 1 , 2 ). Mortality risk was weakly influenced by lower mean critical habitat values within 1 km and was reduced by 18% per 1-unit increase in mean critical habitat (risk ratio: 1.00 – exp(β) = 0.82, SE = 0.21, 95% CI = 0.57, 1.18, Table 2 , Figure 5 and Supplementary Figure 3 ). The top survival model fit reasonably well (Concordance statistic = 0.64; Likelihood Ratio Test: χ² = 7.05, df = 2, p-value = 0.03).

In the Saskatchewan multivariate survival model, we included 52 sage-grouse (12 males and 40 females, n = 1,508 total locations; Table 1 ). Our top survival model ( Tables 1 , 2 ) indicated that mortality risk was strongly influenced by both mean critical habitat and anthropogenic disturbance density within 0.56 km. Mortality risk decreased by 37% per 1-unit increase in anthropogenic disturbance density (risk ratio: 1.00 – exp(β) = 0.63, SE = 0.23, 95% CI = 0.41, 0.97, Table 2 ) and increased by 82% per 1-unit increase in mean critical habitat (risk ratio: exp(β) = 1.82, SE = 0.22, 95% CI = 1.28, 2.59, Table 2 ). The model exhibited good model fit (Concordance statistic = 0.70; Likelihood Ratio Test: χ² = 9.53, df = 2, p-value = 0.01). The survival model and associated risk surface mapped higher risk of mortality in areas with lower densities of anthropogenic disturbance and higher mean critical habitat densities within 0.56 km ( Figure 6 and Supplementary Figure 4 ).

We examined iSSF calibration plots and determined we could fit iSSFs to seven individuals (n = 900 locations) in Alberta and 20 individuals (n = 406 locations) in Saskatchewan. In both provinces, nearly all birds maintained a relatively straight path or made only slight changes in direction when step lengths represented daily steps (negative turn angle; Alberta: β = −0.49 ± 0.31; Saskatchewan: β = −0.53 ± 0.19). In the Alberta iSSF, the selection coefficient for distance to anthropogenic disturbance remained around zero, denoting that at a mean step length (3.23 km), anthropogenic disturbance was neither selected nor avoided by sage-grouse in general ( Figure 5C ). In Alberta, numerous birds selected areas with higher proportions of habitat (mean critical habitat within 1 km; β = 3.93 ± 2.69; Figure 5C ). We also found evidence that sage-grouse made movement decisions that minimize exposure to risky areas compared to what is available to them (β = −8.34 ± 4.08, Figure 7A ). By contrast, in Saskatchewan, a step was more likely to end in greater anthropogenic disturbance density (β = 1.54 ± 0.81) and in greater mean critical habitat both within 0.56 km (β = 2.41 ± 2.49), compared to average values (mean step length = 1.18 km) (Figure 6C). Additionally, the population in Saskatchewan appears to be more likely to select for risky habitat compared to what is available (β = 1.08 ± 1.14, Figure 7B ).

Release sites and surrounding areas differed in terms of their proximate mortality risks ( Figure 8 ). The release sites (30-m pixels) were generally located in sage-grouse habitat with higher pixel scores in the critical habitat layer and were in locations associated with lower mortality risk (i.e., northwest release site in Alberta, Figure 8A ; both release sites in Saskatchewan, Figure 8C ). The southeast release site in Alberta was in a location with lower-scored habitat, with lower mortality risk ( Figure 8B ). However, the areas surrounding release sites included areas of high mortality risk, with patterns and risk levels that differed among sites. For example, the two Alberta release sites had very different mapped risks ( Figures 8A, B ), suggesting that post-release survival could be different between these places. Although we did not have the sample size to compare survival between these two locations, future studies may amass enough data to do so.

We assessed how environmental factors can influence the mortality risk of captive-bred sage-grouse that were released into the wild. Mortality risk of birds that survived longer than 10 days was influenced by anthropogenic disturbance and habitat factors in both Alberta and Saskatchewan, although responses were somewhat nuanced and different between these places. In Alberta, anthropogenic disturbance was the most influential factor influencing mortality risk. Increased exposure to areas proximate to anthropogenic disturbance substantively hastened sage-grouse mortality ( Table 2 ), as demonstrated by elevated risk ratios. As predators often use anthropogenic infrastructure for perching, hunting, or movement (Connelly et al., 2000; Hagen et al., 2011; Baxter et al., 2013), higher mortality in these areas was expected and observed in other studies (Lammers and Collopy, 2007; Prather and Messmer, 2010; Slater and Smith, 2010; Dinkins et al., 2014; Howe et al., 2014; Dinkins et al., 2016). Anthropogenic disturbance can mask approaching predators (e.g., noise from disturbance, Hovick et al., 2014), and can also cause avoidance of areas by sage-grouse, which has been previously documented in southeastern Alberta (e.g., Aldridge and Boyce, 2007). For example, translocated sage-grouse are known to select habitats further from anthropogenic disturbance, up to 10 and 15 km from buildings and settlements, respectively, and up to 2.5 km from roads (Balderson, 2017). Wild sage-grouse in Alberta also avoided oil and gas infrastructure up to 1.9 km from energy development during winter (Carpenter et al., 2010), but did not consistently select habitats that increase survival (Aldridge and Boyce, 2007; Heinrichs et al., 2018). In Alberta, the mortality of wild sage-grouse chicks increased in areas with a higher visible well density within 1 km (Aldridge and Boyce, 2007). We observed a similar mortality response for adult birds released from captivity at the 1-km scale but did not have the sample size to include individual anthropogenic features in the multivariate model. We combined anthropogenic features (i.e., paved and gravel roads, residences, industrial developments, and active wells; Supplementary Table 2 ); hence, our results did not identify the specific built features that most impact sage-grouse survival. However, univariate models indicated distance to gravel roads were more influential than other anthropogenic features ( Supplementary Table 9 ). Future analyses could assess and compare singular anthropogenic features (e.g., only roads) to help managers understand which features have the greatest influence on mortality.

In Saskatchewan, anthropogenic features and habitat were key variables but their relationships to mortality differed from those in Alberta. Contrary to expectation, sage-grouse that used areas with lower densities of anthropogenic disturbance (within 0.56 km) and higher-scored habitat (proportion of mean critical habitat in 0.56 km moving window) were associated with hastened times to mortality ( Table 2 ). This result may be a consequence of the unique environmental context in Saskatchewan. Areas occupied by released, captive-bred sage-grouse tended to be located in the valley bottoms where there is a low density of development, a high proportion of high-selection habitat, and near water resources and higher densities of shrub cover. Predators may be attracted to these productive areas and predation may be high despite the lower availability of anthropogenic features. Thus, mortality may be high in low disturbance areas with high selection habitats because of predators making similar resource choices.

Although environmental risk factors were comparable among release provinces, the differences in mortality risk for sage-grouse underscores that mortality risk can be context dependent. The risk factors were different in each region of release, and these risks were heterogeneously distributed across the landscape, resulting in local and regional maps that could be used as decision criteria in subsequent plans for selecting release sites and regions. Although habitat should be expected to contribute to sage-grouse survival (as observed in Alberta), habitat and associated maps can be an insufficient indicator of mortality risk or of the quality of a release site and its local environment. Habitat maps can provide an initial guess of where to release captively-bred animals; however, these guesses may be misleading when other environmental factors that influence survival are not included and mapped in habitat analyses. We included environmental variables that represent opportunities for predation, including anthropogenic and tall features, topography, and restricted cover, but lacked spatial information that described predator abundance and distribution. When available, this information could be used to augment spatial risk analyses and help further identify the potential causes of mismatches in habitat and survival conditions (e.g., O’Neil et al., 2020).

We found mixed evidence that released sage-grouse respond to environmental conditions in a way that minimizes mortality risk. The step-selection function results for sage-grouse released in Alberta suggest they generally moved and selected locations that were in less risky habitat compared to what is available to them ( Figure 7A ). They selected for higher-scored habitats (greater mean critical habitat; Figure 5C ), further away from anthropogenic disturbance (although not significant; Figure 5C ) and these choices resulted in a slower or decelerated time to mortality ( Table 2 ; Figure 5 ). Avoidance of anthropogenic features is consistent with how wild sage-grouse select habitats within various critical life history stages (e.g., nesting, brood-rearing, Aldridge and Boyce, 2007) and seasons (e.g., breeding, summer, winter, Gelling, 2022). However, sage-grouse do not always optimize fitness with habitat selection (e.g., maladaptive selection; Aldridge and Boyce, 2007; Cutting et al., 2019; Lazenby et al., 2020; Pratt and Beck, 2021). Wild birds can select familiar places (e.g., nesting sites) that have lower nest or chick survival (Aldridge and Boyce, 2007), resulting in population sources and sinks (Heinrichs et al., 2018). Accordingly, the movement selection results for Saskatchewan indicated that released grouse selected risky areas. Their preferences for resource-rich sites may have outweighed their perceived risk of predation. This inconsistency in released birds’ movement and selection behaviors among release areas emphasizes the value of building models for unique environmental contexts. When sufficient data becomes available to assess mortality risks in specific areas, we recommend updating model extrapolations from other areas with site-specific models.

We lacked sufficient data to fully assess mortality risk using step selection functions. However, we found this multistep approach to be a useful coupling of methods to assess congruence of survival with movement selection and avoidance; ultimately, resulting in evidence that released individuals are making at least some choices that optimize their fitness, and some that are not. We used daily sage-grouse locations, but with increased temporal resolution of the data, this analysis could be expanded to assess finer-scale movement behavior that may further illuminate risks and movement responses associated with specific environmental features.

Coyotes and raptors, such as the great horned owls, are the apex predators of this ecosystem (Ritchie and Johnson, 2009) and use anthropogenic infrastructure to move and hunt. We lacked data on predator abundance and distribution and were unable to directly assess risk factors associated with predation; however, as most sage-grouse deaths were caused by predation, we infer mortality risk is primarily driven by predation and factors that facilitate predation. High mortalities in the presence and absence of anthropogenic infrastructure (in Alberta and Saskatchewan, respectively) indicate there is more to learn about the causes of mortalities in each place, including predator communities and the environments that facilitate predation. This risk assessment could be augmented with maps that describe predation pressure and predator abundance and distribution results from long-term systematic surveys. Mortality risk maps (e.g., Figures 5A , 6A ; Supplementary Figures 3, 4 ) could be used to support the design of predator surveys and utilization maps to spatially target areas in which to deter predator use (e.g., through removal of attractants; Parks Canada Agency, 2021). This may be particularly useful for canid predators, such as coyotes, that are opportunistic feeders and able to exploit a variety of environments and resources (Bartel and Knowlton, 2005).

Further analyses of predator data can be strengthened with molecular analyses to improve predator determination (Peelle et al., 2019); however, preliminary analyses suggest that canids, corvids (e.g., American crows, Corvus brachyrynchos; common raven, Corvus corax) and owls (e.g., great horned owl) (Quinlan, 2013) are key consumers of released sage-grouse. The expansion of corvid surveys in Alberta to include a broader range of predators would help researchers and managers better understand sage-grouse mortality risk. Future analyses could also consider how the landscape facilitates and protects sage-grouse from weather exposure during extreme weather situations, as well as incorporate additional risk factors as data become available (e.g., smaller built features such as power poles, transmission lines, fences).

Beyond environmental associations of mortality risk, we also examined the factors associated with release including age and weight at release, acclimation period, raising method, release period (e.g., release before October 17, or release after October 17), and release type (e.g., hard, soft, and modified-soft release). Our results suggest the spatial-environmental factors that we measured were generally more influential than factors related to the release protocol in affecting overall mortality risk. However, we did find that some of these protocol-related factors were explanatory, depending on how the data were analyzed, for which we sought alternative ways to analyze the data (see Supplementary Material for alternative models). For example, if we looked at mortality outcomes by season, the model fit was not as strong, but we occasionally found a signal of age and weight at release (only in Saskatchewan during the fall; Supplementary Figure 2B ).

Although the influence of animal traits and release protocol factors were minor relative to the release environment, there could be other unquantified conditions that influence mortality including animal health (e.g., stress), predator avoidance conditioning, and diet adaptation. For example, some birds can die earlier due to health or disease-related issues and (or) because of lack of environmental conditioning (e.g., diet is the ultimate cause, and predation is the proximate cause). Captive breeding and translocation programs can impose stress on animals during captivity, handling, transportation, and release experiences (Dickens et al., 2010). Stress can influence behavior (e.g., risk taking; May et al., 2016) including movement and habitat selection (Osborne and Seddon, 2012), and reduce survival of reintroduced individuals (Bremner-Harrison et al., 2004). Stress and disorientation induced by the release can lead individuals to select less suitable areas outside of strategically chosen release sites with suitable habitat. Individuals may have further difficulties in transitioning between supplementary feed and natural diets due to their reliance on supplementary feeding (Draycott et al., 1998). Survival times are generally longer for wild-to-wild sage-grouse translocations in Alberta, with average hen survival (n = 77 with known fates) of close to a year (296 days, and some survived >1 year; unpublished data), suggesting that there are likely to be some unquantified factors beyond environmental features that are important to the survival of released sage-grouse. As these factors are quantified, they can be used to enrich this analysis.

The results of this analysis can support adaptive planning and management of the release program and environment. The success of captive breeding and translocation programs depends not only on the quantity and quality of animals released, but also on identifying and managing the factors that contribute to their survival in the wild (Berger-Tal et al., 2020). Some of the elements associated with post-release mortality can be managed, reduced, or mitigated. Beyond optimizing captive and release protocols, the environment into which the animals are released should also be managed to reduce mortality risk (IUCN/SSC, 2013). For example, mortality risks for Galliformes can be partially mitigated by selecting release locations and surrounding areas that minimize opportunities for predation or predation pressure (Warren and Baines, 2018), and maximize food and cover resources. Future release sites and surrounding areas in Alberta and Saskatchewan can be screened using the risk map or associated spatial data layers to further optimize site selection. Sage-grouse release areas differed in their habitat characteristics and mortality risks. Thus, future release sites could be chosen to balance the risk profile of existing sites by releasing animals into areas with lower overall risk and a different balance of risk factors. This cycle of adaptive management can benefit from strong collaborations among translocation professionals and data scientists that can make use of small datasets, as well as species and land managers and stewards that can put results into action.

Actions are being taken by a range of government, conservation, and industry groups to reduce anthropogenic infrastructure in Canada and benefit sage-grouse. Results suggest the removal of oil and gas infrastructure, and predator perches including trees, buildings, and other vertical features (Whiklo and Nicholson, 2015) will change the spatial distribution of mortality risk, and these actions have the potential to reduce mortality risk for released sage-grouse. However, it is unclear whether infrastructure removal by itself would be sufficient to sustain released sage-grouse. As wild populations have been declining for decades in Canada, it is likely that several factors are contributing to high mortality rates, and several kinds of population recovery and habitat restoration actions are needed to stabilize the population trajectory. Although some threats sage-grouse face cannot be directly managed (e.g., climate patterns, severe weather events), a number of threats can be mitigated indirectly through habitat improvement such as encouraging natural hydrology in mesic areas (e.g., constructing ditch or dam) to augment sage-grouse forage (Alberta Environment and Sustainable Resource Devlopment, 2013; Environment Canada, 2014). Models that link spatial mortality risk to landscape projections could be used to scope the population gains associated with complementary actions that benefit both wild and released sage-grouse (Heinrichs et al., 2018, Heinrichs et al., 2019).

We used a multiphase approach to assess the environmental factors influencing mortality risk, whether existing habitat maps are likely to predict where released animals are likely to survive, and the degree to which animals’ movements corresponded to risk correlates. We implemented a workflow that addresses the challenge of having few surviving animals and small location datasets, a situation common in captive breeding and release programs. We suggest this linkage of analyses is helpful for several reasons. First, it yielded a simple yet actionable risk map that complements existing information on habitat, populations, and landscape features. Viewing habitat and mortality risk in the same map highlighted locations where the conditions are appropriate for both habitat selection and mortality, and locations where landscape improvement actions may yield high returns on ecological investments. Second, this workflow was also useful in evaluating the assumption that ‘good’ habitat in habitat selection maps represent optimal places in which to release translocated animals. We found several locations where highly selected habitat is associated with a high risk of mortality. Mortality risk maps can complement existing habitat maps by including additional factors that were excluded from the habitat map that may not have a strong signal of selection but influence survival. Our mortality model included anthropogenic and other features that facilitate predation that were not included in the habitat model. Mortality models could additionally include predator locations, densities, preferences, or locations that lack resources to enable animals to withstand extreme weather. Lastly, this workflow evaluated movement behavior relative to the risk landscape and provided a means to assess whether animals are responding to risk features. Movement and selection responses were different between release areas (Alberta and Saskatchewan), indicating that release success, and the prescriptions to enhance post-release survival, may differ among locations. Actions that address high mortality in high selection areas may be more influential than those in low selection areas (e.g., Heinrichs et al., 2018).

Although we found a path to mapping post-release mortality risk that led to key management insights, this approach had limitations. For instance, we used a small dataset where seasons, sexes, and ages were combined, and environmental covariates were combined in composite layers. Consequently, we may have missed relationships that could be found in larger datasets, including those related to release protocols and interacting environmental conditions. A larger dataset would have streamlined the workflow by allowing us to jointly evaluate several environmental covariates in a single model, as well as investigate differences among individuals (e.g., traits, status), differences in release protocol, and differences among behavioral and movement states (e.g., encamped, exploring, travelling, relocating; Harju et al., 2013; Picardi et al., 2021). Despite these limitations, we found empirical patterns that can be used to support near-term conservation actions while additional data are collected and used to develop the next generation of risk models. We used concepts and tools that most wildlife researchers are familiar with, using free and accessible software packages. As such, we suggest this workflow may be useful for other wildlife researchers facing similar constraints and may be particularly useful in cases where mismatches in habitat selection and survival are likely to occur.