INTRODUCTION

Targeted monitoring of keystone and endangered species in an ecosystem is critical for their successful management and recovery, and has the best chance at success when management actions are taken before populations have declined drastically (Taylor et al. 2005, Nichols and Williams 2006). Monitoring programs capable of detecting early warnings of species declines are therefore needed, yet monitoring can be particularly challenging in northern environments like the boreal forests of Canada because of the low number of individuals across vast, remote areas (Allen 1952, Ausband et al. 2014). In this case, species distribution modeling is particularly useful to make inferences across large areas and detect changes in estimated species occurrences (Van Wilgenburg et al. 2020, Beirne et al. 2021, Bobrowski et al. 2021). It is important that the sampling methods chosen for data collection are effective in meeting these objectives within the context of a given study area and monitoring program (Hanson et al. 2023).

Scientists are often under pressure to choose data collection methods that will not only provide accurate and precise information, but also be less invasive, more cost-effective, and more in line with local Indigenous values (Soulsbury et al. 2020, Wong et al. 2020, Hanson et al. 2023). However, many common wildlife survey techniques do not meet these criteria. For example, aerial surveys can be problematic because they can disturb the target species (Bleich et al. 1994, Ditmer et al. 2015, Brambilla and Brivio 2018) and can result in significant under-estimates of population sizes due to low sightability (Fleming and Tracey 2008, Schlossberg et al. 2016). Methods that involve handling animals, such as telemetry, may be against cultural values of Indigenous peoples (Spak 2005, Wong et al. 2020). This is particularly important in northern Canada where wildlife management is generally a shared responsibility between federal, territorial, and Indigenous governments and co-management boards established under land claims (White 2020). Furthermore, the more expensive and labor-intensive, tracking, tagging, or marking animals can also result in changes in behavior or reduced fitness (Northrup et al. 2014, Soulsbury et al. 2020). Even call-response surveys, which broadcast recorded animal calls to elicit a response, may disrupt natural behavior and potentially violate assumptions made when modeling those data (Conway and Gibbs 2005, see also the use of baited stations, Stewart et al. 2019). Although these sampling methods provide important information on species distribution, abundance, and population dynamics, there remains a need to complement these conventional methods with new methods that tackle these challenges.

Camera traps (hereafter cameras) and autonomous recording units (ARUs) are both cost-effective and minimally invasive sampling methods to collect wildlife data (Burton et al. 2015, Shonfield and Bayne 2017). They can be deployed at the same time, used for targeted studies and long-term monitoring of diverse taxa, and integrated with other sampling methods. For example, ARU surveys can be integrated with in-person avian point count surveys for a more comprehensive sampling design (Van Wilgenburg et al. 2017, Drake et al. 2021). Deployments of cameras and ARUs can also be combined (Buxton et al. 2018) to monitor a wider range of species and ecological processes, and leveraging funding across a wider range of collaborators with shared priorities (ABMI 2023, Environment and Natural Resources 2023). Autonomous recording units are widely used to study birds as well as many other taxa (Barber-Meyer et al. 2020, Palacios et al. 2022). Similarly, cameras are widely used to study mammals and have been shown to effectively detect some birds (Fontúrbel et al. 2020, Beirne et al. 2021, Sun et al. 2021, Vaughan et al. 2022). Working with multiple sensors across multiple taxa can bring together many collaborators, which increases the resources and expertise available to a given project. One example of large-scale, multi-taxa monitoring using cameras and ARUs (among other methods) and involving collaborating groups is the Alberta Biodiversity Monitoring Institute (Burton et al. 2014; https://abmi.ca/abmi-home).

Despite the ease of pairing cameras and ARUs, there has been limited research on integrating the two types of data in a single modeling framework. Studies have compared both sampling methods for occupancy and detectability of gray wolves (Garland et al. 2020), seasonal distribution and detectability of right whales (Rayment et al. 2018), diversity of understorey birds (Fontúrbel et al. 2020), detectability of hoary bats (Gorresen et al. 2017), and relative abundance of penguins (Francomano et al. 2024). Most of these studies concluded that either one method or the other was superior from a stand-alone point of view, but that detection probabilities were highest when both sensors were deployed together. Autonomous recording units and cameras have also been paired to examine the effects of anthropogenic disturbance on animals: for example, poaching and livestock (Dobbins et al. 2020, Vélez et al. 2024), and the effects of anthropogenic noise on seabirds and mammals (Buxton et al. 2017, Kleist et al. 2020). And separately, both methods have been used to build species-habitat models for conservation planning (Shonfield and Bayne 2017, Kays et al. 2024). With increasing interest in combining data from different sources through integrated models (e.g., Miller et al. 2019, Isaac et al. 2020), there is an opportunity to test whether an integrated habitat modeling framework using both camera and ARU data could improve on habitat models based on only one data source.

Habitat selection is a hierarchical process that occurs across different spatial scales (Johnson 1980, Wiens 1989, Toews et al. 2017). Therefore, the scale of measurement, in terms of both spatial resolution and extent, should match the spatial scale of inference and analyses should be conducted at multiple spatial scales for a better understanding of these processes (Mayor et al. 2009, Wheatley and Johnson 2009, Toews et al. 2017). Inferences made based on ARU and camera data might reflect habitat associations at different spatial scales because ARUs can typically detect vocalizations at greater distances than the field of view of cameras, especially in forested environments (Sólymos et al. 2013, Becker et al. 2022). Integrating these methods may thus allow for multiple scales to be analyzed at once.

Sandhill Cranes (Antigone canadensis; Dǝle in K’áshógotįne Goxədǝ́) is an ideal species to test the usefulness of integrating ARU and camera data because this species is easily detected by either sensor (Fig. 1). Sandhill Cranes are very vocal; they advertise their breeding territory with rattle calls that can be heard up to four kilometers away (Gerber et al. 2020), and as such are well-detected by ARUs. Sandhill Cranes even possess tracheal elongation, which causes their calls to be louder than expected for their body size (Fitch 1999). They are also large-bodied and travel almost exclusively on the ground before their chicks are able to fly (Gerber et al. 2020), making them a good candidate for monitoring using camera traps. In Canada, Sandhill Cranes are harvested (Government of Canada 1994), and monitoring of relative abundance over time could inform adaptive harvest management through regulations such as bag limits, timing, and length of hunting season, etc. A framework to generate habitat models for Sandhill Cranes that would include both ARU and camera data could be applicable to many other large-bodied birds and mammals to help inform status assessments and conservation actions, including the endangered Whooping Crane (Grus americana; Government of Canada 2002, IUCN 2022).

In this study, we compared predictive performance of different species habitat models using ARU and camera data to better understand habitat associations and distributions of Sandhill Cranes nesting in T’sudé Nilįné Tuyeta Indigenous and Territorial Protected Area, Northwest Territories, Canada. We tested whether (1) inferences about habitat relationships would be different using an ARU data model, camera data model, or integrated ARU and camera data model; and (2) whether an integrated model would have better predictive performance than ARU and camera models alone. Models were built using land cover variables at two spatial scales representing the home range (300 m radius) and a broader landscape context (2000 m radius). Our first prediction was that cameras would be more informative (e.g., explain more variation in detections) at a smaller spatial scale than ARUs because they have a smaller effective detection distance (Becker et al. 2022), and therefore more effectively pinpoint the locations of cranes within home ranges, while ARUs would be more informative about habitat use at a larger spatial scale because of the loud crane vocalizations. Our second prediction was that the integrated model would have the best predictive power given we anticipated that including more data would provide more robust information. We used our results to provide general recommendations for sampling strategies that use both methods.

METHODS

Study area and survey design

Ts’udé Nilįné Tuyeta (hereafter Tuyeta) is a remote 10,060 km² Indigenous and Territorial Protected Area in the Northwest Territories, Canada (Fig. 2). Its climate is subarctic, with a mean annual temperature of -6 °C and annual precipitation of 200 to 600 mm, about half of which falls as snow. It is mainly within the Taiga Plains ecozone, characterized by a low elevational gradient and permafrost (Ecological Stratification Working Group 1995). This area contains a patchwork of upland and lowland habitat, uplands being dominated by black spruce (Picea mariana), Jack pine (Pinus banksiana), and birch (Betula papyrifera), and lowlands dominated by Sphagnum bogs and sedge marshes. In the southern portion of Tuyeta, it transitions through foothills to the MacKenzie Mountain range in the Taiga Cordillera ecozone.

Cameras and ARUs were deployed in pairs at 250 locations throughout Tuyeta (hereafter, paired sensors). To accommodate the priorities of multiple partners in this collaborative monitoring program, paired sensors were deployed using a combination of three sampling designs. First, 157 locations were selected by the Boreal Optimal Sampling Strategy, a boreal-wide stratified random sampling design for migratory birds (Van Wilgenburg et al. 2020). Second, 66 locations were re-surveys of historical avian survey locations (Canadian Wildlife Service 2007). Third, 27 locations were targeted along 2 culturally important river basins (Ramparts and Hume rivers). Sensors were deployed in March 2020 for the first two designs and in July 2020 for the third. Paired sensors were left in place for one year. Sensor locations were clustered to increase sampling efficiency and allow inferences at hierarchical spatial scales (see Van Wilgenburg et al. 2020). A cluster consisted of paired sensors within a five km (in diameter) hexagon for consistency with the Boreal Optimal Sampling Strategy protocol (Van Wilgenburg et al. 2020). Due to in-field challenges, including constraints on accessibility from deep snow and lack of helicopter landing sites, not all planned locations could be deployed, leading to more variation in sensors per cluster than planned. A total of 44 clusters with 1 to 19 paired sensors (5.6 ± 6.2 SD) were deployed across the study area. Within a cluster, paired sensors were between 150 m to 2700 m apart (330 m ± 172).

We used Reconyx Hyperfire HP2X camera traps and Wildlife Acoustics Song Meter SM4 and Song Meter Mini ARUs. Cameras were attached to trees one m above the ground, angled slightly down, facing open viewsheds when available, and preferentially facing north to reduce sun glare (Meek et al. 2016). Cameras were programmed to take three pictures per motion trigger, with no quiet period, high sensitivity, and to take a daily time lapse image at noon to evaluate whether the camera was functioning. Autonomous recording units were programmed to take 10 or more 3-minute recordings per day at dawn, dusk, and night periods throughout the spring and summer. Throughout a portion of the summer in Tuyeta, the sun does not set and there is no period of darkness or twilight; during that time, recordings were made at the times of day when the sun was lowest in its altitude. The ARUs recorded in stereo WAV file format, with a sampling rate of 44100 Hz and factory default settings for the microphone preamplifier.

Image processing and audio transcription

For camera trap image processing, we used Microsoft Megadetector AI v4 (Microsoft 2020) with a conservative confidence threshold of 0.2 to pre-filter blank images. Trained staff reviewed the remaining images to identify animals using the WildTrax platform (https://wildtrax.ca). All species identifications were verified by another observer. Staff also assessed whether the field of view of each image was unacceptable to capture wildlife (i.e., camera lens snowed over or underwater). Survey effort for each camera was calculated by the amount of time, in decimal days, the camera was active and taking acceptable images.

Eight three-minute recordings were transcribed per ARU. All recordings were in the month of June to target the peak of singing activity for most bird species. Although Sandhill Cranes are present in Tuyeta beyond the end of July, we did not include any later times in our analysis because chicks will fledge in August and family groups will begin to abandon their territories (Gerber et al. 2020). Most ARUs (142) recorded in 2020, but 18 ARUs recorded in 2021. For each ARU, four recordings each were randomly selected in early (1 to 15) and late (16 to 31) June. The recordings were randomly selected to represent early morning (0200–0459), mid-morning (0500–0659), late morning (0700–0859), and night (2030–0130). In cases in which the selected recording was unsuitable due to bad weather (rain or wind), another randomly selected suitable recording was used from the same time and date categories or, if not suitable, within the same time category, but in the other date category (i.e., early or late June). However, we never selected two recordings from the same day; because of this limitation, there were some cases in which fewer than eight recordings were transcribed per ARU. We used the hour of the day to categorize the recordings instead of time relative to sunrise because the sun does not set for most of the month of June in Tuyeta; details of sampling effort are shown in Appendix 1, Fig. A1.4. Transcription was done on the WildTrax platform by one human observer who classified all species in each recording.

Of the 250 locations with paired sensors, we obtained data from our target period (May through July for cameras, June for ARUs) for 229 cameras and 160 ARUs, representing 246 unique locations. Causes of camera failure included cold temperature malfunctions and reaching battery and/or memory capacity. Autonomous recording units experienced a high failure rate due to firmware malfunctions in extremely cold winter temperatures. We were able to transcribe the full 8 recordings for 154 of the 160 ARUs, and the remaining 6 ARUs had at least 2 recordings transcribed. Cameras were active and taking acceptable images for an average of 84 days (range = 5–90) throughout our target period. Total sampling effort was 19,236 camera-days and 3789 ARU-minutes.

Habitat model

From both sensors, we used detection-nondetection data to estimate Sandhill Crane occupancy, i.e., the probability that a crane occurs at a given location (MacKenzie et al. 2017, Stiffler et al. 2018, Wevers et al. 2021). We fit occupancy models with the package spOccupancy (Doser et al. 2022) in R statistical software (R Core Team 2022). This modeling framework assumes that the true presence (1) or absence (0) of a species follows a Bernoulli process described by the occupancy probability (ψ). Occupancy probability itself is modeled based on the occupancy covariates and estimated coefficients, using a logit link function. The model also assumes that the detection (1) or nondetection (0) of a species across multiple visits at the same sampling location also follows a Bernoulli process, described by the detection probability (p), conditional on the true occupancy at a given location. Detection probability is modeled based on visit-specific detection covariates and coefficients, also following a logit link function. This modeling framework is also capable of integrating multiple data types; it does this by modeling detection probability separately for each data source and modeling occupancy probability jointly (MacKenzie et al. 2017, Doser et al. 2022).

There are five key assumptions to occupancy models (Mackenzie et al. 2017):

1. locations are “closed” throughout the duration of a sampling period, i.e., there are no changes in species occupancy between visits;
2. locations and visits are independent;
3. there are no false detections (no false positives);
4. detection probability is constant across locations, or modeled adequately with covariates; and
5. occupancy probability is constant across locations, or modeled adequately with covariates.

In practice, some of these assumptions may be violated, and there is ongoing research on the consequences of potential violations (e.g., Neilson et al. 2018). Nevertheless, we assumed that our data adequately met assumptions. For the closure assumption (1 above), sampling locations approximately corresponded to crane home ranges, yet crane movements may have violated the assumption; accordingly, we followed the commonly used interpretation that movements in or out of a site were random with respect to sampling, and that “occupancy” is more appropriately interpreted as “use” (Mackenzie et al. 2017). Further, detections may not have been independent among sampling locations within a cluster (assumption 2), which motivated our use of a random effect for cluster-level variation, but we acknowledge that future research could evaluate the potential for spatial autocorrelation through further model complexity (e.g., spatial autocovariate).

Detection histories

We summarized binary detection-nondetection of Sandhill Cranes for each visit at each location in the form of detection histories. For ARUs, detection histories were detection or nondetection on each three-minute recording. For camera data, we used weekly detection histories throughout the target period. Although there is a difference in the sampling period between the two methods, the assumption of closure should hold throughout this period because Sandhill Crane pairs successfully raising chicks will typically occupy the same territory throughout the breeding season (Gerber et al. 2020). Furthermore, it may be necessary for cameras and ARUs on Sandhill Cranes’ breeding grounds to target different times within the cranes’ phenology because peak vocalization occurs before incubation, while peak movement rate occurs once the chicks have hatched (Gerber et al. 2020). Thus, our sampling periods reflect appropriate sampling conditions for each method.

Occupancy and detection covariates

To test whether occupancy of cranes at a sampling location was explained by habitat characteristics, we extracted habitat variables using two buffer sizes centered on each ARU or camera. The territories of Sandhill Cranes are relatively small, ranging from 0.17 to 1.4 km², depending on the study area (Drewien 1973, Walkinshaw 1973, Reed 1988); therefore, we chose 300 m to represent habitat selection for a typical home range of a pair of Sandhill Cranes. Despite their small territories, Sandhill Crane calls are extremely loud; while a specific effective detection radius (which describes the relationship between detection probability and distance; Matsuoka et al. 2012) has not been published, it is believed that their calls can be heard up to four km away (Gerber et al. 2020). A maximum detection distance is typically twice (or more) the size of the effective detection radius (Buckland et al. 2001), so we reasoned it would be approximately two km, and therefore used this size for the larger buffer. Recognizing that this will be variable based on external factors (e.g., weather, vegetation), we considered this buffer to represent habitat selection within a larger landscape context, and we refer to it as the landscape scale for simplicity.

For occupancy covariates, we used marsh, fen, bog, swamp, and total wetland areas based on the Ducks Unlimited Canada wetland classifications (Smith et al. 2007, Ducks Unlimited Canada 2021). We focused on the four wetland land-cover types given previously reported habitat association for this species (Saunders et al. 2019, Gerber et al. 2020). We calculated percent cover of each land-cover class within both buffers for each location and scaled them (i.e., subtracted mean and divided by one standard deviation). To reduce multicollinearity, we did not include variables with a correlation of greater than 0.6 in the same model (Dormann et al. 2013), and therefore did not include fen and marsh or fen and swamp in the same model at the landscape scale (Appendix 1, Fig. A1.3). We used cluster as a random effect to account for the non-independence of repeated measurements within the cluster (i.e., spatial proximity of sampling locations).

Detection covariates for ARU data were Julian day, Julian day squared, and time of day. We chose these covariates because there are peaks in Sandhill Crane detectability (i.e., peak time of vocalization) throughout the month of June and early in the morning (Gerber et al. 2020). Detection covariates for camera data were the number of active camera days within each week (i.e., survey effort), week of the year, and week of the year squared. Number of active camera days within each week would directly affect detection probability, and there is a peak in movement rates that would affect detectability within the period we surveyed.

Model specifications and selection

We fit models using noninformative priors and default initial values for regression coefficients. Specifically, priors were set with hypermeans of 0 and hypervariances of 2.72 for both occupancy and detection covariates (Lunn et al. 2013). We began with 3 chains each of 5000 Markhov Chain Monte Carlo (MCMC) samples with a burn-in of 1000 samples and a thinning rate of 2. We increased the number of samples until the Gelman-Rubin diagnostic was 1.1 or less for all estimated coefficients, indicating convergence (Brooks and Gelman 1998). We also performed posterior predictive checks as an assessment of goodness of fit using the Freeman-Tukey fit statistic (Kéry and Royle 2016).

To find the most parsimonious model for each data type, we ran multiple models and chose the top-ranking one using the widely applicable information criterion (WAIC; Hooten and Hobbs 2015). Because of computational limitations, model selection was iterative. First, we ran 7 competing models for each data type (ARU, camera, and integrated): all wetland types except fen (due to multicollinearity l), fen and bog, each of the 4 wetland types separately, and all wetland types added together, using the 2000 m buffer. We then chose the two top-performing models from this first pass of model selection and ran those models with all permutations of the corresponding covariates and spatial scales. We then compared all resulting models and chose one top-performing model for each data type.

For the integrated model, we integrated both data types and spatial scales. We reduced multicollinearity between percent cover at each buffer size by using a similar approach described by Rhodes et al. (2009). In short, we performed a linear regression between the same habitat type at both buffer sizes, using percent cover within home range scale as the x-variable and percent cover with the landscape scale as the y-variable. Each individual location with one pair of sensors (n = 246) was one data point. We then calculated the residual value for each point (Appendix 1, Fig. A1.1), which represents whether there was more or less cover of each habitat type at the landscape scale than would be expected based on the home range value. In our models, we included the value for the home range scale and the residual value.

We used four-fold cross-validation (Hooten and Hobbs 2015) to examine the predictive performance of the integrated model relative to the camera only and ARU only models. Cross-validation can be used to compare the predictive performance of different models, but only if the testing data comes from the same dataset (Hooten and Hobbs 2015). Therefore, we couldn’t directly compare the predictive performance of the top-ranking camera and ARU models. However, we assessed whether the addition of each data type in the integrated model improved predictive performance compared to using one data source alone. We first cross-validated the top-ranking ARU model, choosing testing data randomly from all locations and fitting the model to the remaining data. We compared the resulting deviance score to cross-validation done on the top-ranking integrated model, in which we chose testing data randomly only from the ARU dataset, and fit the model with the remaining ARU data plus all camera data. We repeated this process with the top-ranking camera model and compared that score to cross-validation done on the top integrated model, where we held out camera data only for testing.

For each data source, we compared the size and precision of the detection and occupancy parameter estimates. We calculated an overall detection probability for each data source using p* = 1 - (1 - p)^k, where p is the detection probability for a single visit, and p* is the probability of detecting a crane at least once across k visits, given that the location is occupied (MacKenzie et al. 2017). For k, we used the average number of surveys each data source completed, which was k = 8 recordings for ARUs and k = 12 weeks for cameras. To visualize how inferences would change depending on the data source, we also mapped predicted occupancy across Tuyeta based on the top-performing model from each data source. To do this, we created a 1 km by 1 km grid of points across the study area. At each point, we extracted percent cover of wetland land-cover classes within a 300 m and 2000 m buffer. Based on that information, we predicted point-level occupancy using the coefficient estimates from the top-performing ARU model, camera model, and integrated mod el, respectively, keeping detectability at a mean value. We then examined each map for differences and similarities in patterns of occupancy probability. Finally, we calculated the Pearson correlation coefficients among predictions made based on each data source.

RESULTS

Sandhill Cranes were detected on 59% of ARUs, 18% of cameras, and 48% of the total locations (naive occupancy). Of the 143 locations where data were obtained from both ARUs and cameras, cranes were detected 63 times by ARUs only, 17 times by both methods, and 9 times by cameras only. Posterior predictive checks resulted in Bayesian p-values of 0.45 for the top ARU model, 0.12 for the top camera model, and 0.32 and 0.00 for the ARU and camera portions of the integrated model, respectively (Appendix 1, Fig. A1.2). The ideal p-value is 0.5, which indicates that the predicted and observed values match.

All top models for occupancy included percent cover of marsh, swamp, and bog, with fen also included in the camera model (Table 1). As expected, the top ARU model included habitat variables at the 2000 m buffer, while the top camera model included the 300 m buffer, and the top integrated model included variables from both spatial scales (Table 1). The top ARU model was only slightly better than the null model (ΔWAIC = 1.17), but the top camera and integrated models were notably better than the null models (ΔWAIC = 37.26 and ΔWAIC = 53.39, respectively).

Sandhill Crane occupancy increased with percent cover of marsh and bog, except within the 2000 m residual for cover of bog within the integrated model. The effect of swamp on occupancy was always negative, but did overlap zero in the ARU and camera only models (Fig. 3). The camera model had the lowest intercept, indicating lowest average Sandhill Crane occupancy, and the intercept was highest for the integrated model. Marsh and swamp were included in the ARU model, all four wetland classes were included in the camera model, and bog, swamp, and marsh were included in the integrated model.

Detection probability was also lower for the camera model than for the ARU model (Fig. 4). For ARU data, detection probability was p = 0.22 and detection probability across all repeated visits was p* = 0.87. For camera data, p = 0.08 and p* = 0.62. Highest detection probability for ARUs was found to be earlier in June (negative value for Julian day) and later in the day (positive value for time of day), while highest detection probability for cameras was found to be later in the summer period (positive value for week of the year) and on cameras with more active days per week (positive value for active days; Fig 4.). The integrated model estimated very similar detection coefficients, but with slightly lower intercepts for both ARU and camera data; the estimated detection probability for ARUs was p = 0.22 and p* = 0.86, and the estimated detection probability for cameras was p = 0.03 and p* = 0.31.

Surprisingly, the deviance score for the camera model was higher than the corresponding integrated model, but the deviance score for the ARU model was lower (Table 2). In other words, including ARU data in the integrated model improved predictive performance for the camera data, but including camera data in the integrated model did not improve predictive performance for ARU data.

The predicted occupancy for the top ARU, camera, and integrated models show similar patterns (Fig. 5). The camera model predicts smaller areas of high occupancy, while the ARU and integrated models predict overall higher occupancy across the study area. Across all points, the correlations among predicted occupancy were 0.31 for ARU and camera, 0.88 for ARU and integrated, and 0.46 for the camera and integrated models.

DISCUSSION

Across data types and models, our results show that Sandhill Crane occupancy was positively associated with the percent cover of marsh, fen, and bog around a location and negatively associated with the percent cover of swamp. Positive association with wetland habitat types, particularly fen and bog, is in accordance with previous studies of Sandhill Cranes (Melvin et al. 1990, Baker et al. 1995, Kruse et al. 2017). However, a study by Kruse et al. (2017) during nesting season in Colorado, USA, showed positive associations with swamp. This discrepancy may be explained by how land cover is characterized by different wetland classification systems; the U.S. National Vegetation Classification (2016) includes peatlands in their definition of swamp, which may be classified as bog by the Ducks Unlimited Wetland Classification (Smith et al. 2007). Despite the general consistency in habitat associations, our methodological comparison highlighted important differences between data types and associated models.

There are two key differences in the detection process between ARUs and cameras: the duration of sampling, which is much larger for cameras, and the effective area sampled, which is much larger for ARUs. Only 8 3-minute recordings were transcribed per ARU, for a total of 24 minutes of sampling each, while cameras were active around the clock for an average of 84 days (5000 times longer sampling period). Given an estimated effective detection radius of 2 km for the call of the Sandhill Crane, the effective area sampled for ARUs would be approximately 12.5 km² (Sólymos et al. 2013). The effective area surveyed by the camera, however, could be estimated as ~10 m in front of the camera in a 40° arc, equaling 125 m² (Becker et al. 2022). The effective area surveyed is a key distinction, especially for occupancy models, because the definition of occupancy is dependent on scale (MacKenzie et al. 2017). For the same number of individuals across a study area, choosing a larger scale to represent one site will result in a greater estimation of occupancy (Miller et al. 2019). This may partly explain the higher occupancy estimate derived from ARUs versus cameras.

The loudness and frequency of the Sandhill Crane vocalizations make this species more likely to be detected on ARUs than on cameras. In fact, Sandhill Cranes were so common on ARU data that the models were more difficult to fit; they required many more iterations (200,000 versus 50,000) to reach convergence, and the WAIC value of the top model was closer to the null model than it was for the camera model. However, the Bayesian p-value for the ARU model was closest to 0.5, indicating the best model fit of the three data sources. Work by Baker et al. (1995) found no significant habitat associations beyond a 200 m buffer around Sandhill Crane nests in Michigan, USA and recommended the scale of habitat selection should be carefully considered for this species. This could be a concern for habitat relationships derived from ARUs because an ARU that detects a crane might actually fall outside of its home range and preferred habitat type; ARUs may also detect multiple pairs of cranes occupying different home ranges. Nonetheless, a larger buffer size would still encompass the home range of a crane that was detected, and a higher proportion of preferred habitat within the buffer would still lead to a higher probability of occupancy. The ARU models did provide information on habitat selection of cranes because our top model was slightly better than the null model. However, some precision may be lost by the spatial scale mismatch. More work is needed to better understand the proper spatial scale of inference when using ARU data to model occupancy of Sandhill Crane and ecologically similar species.

In contrast to the ARU results, the top-performing camera model estimated lower overall occupancy and included covariates at the smaller buffer size (300 m). These differences suggest that cameras measure occupancy at a smaller scale and reflect habitat selection within home ranges. Camera data show the exact location of crane adults and offspring, as well as their behavior (Fig. 6). The occupancy map predicted from camera data is patchier and areas of higher occupancy should be interpreted as areas of very high-quality habitat. Given the different spatial scales of detection for the two sampling methods, it was not surprising that the camera model estimated lower occupancy than the ARU model. The pattern of the camera map is generally similar to the pattern of the ARU map for areas of high-quality habitat, indicating general agreement between the two methods. The Bayesian p-value of the top camera model is relatively low but above 0.1, indicating an acceptable model fit (Hobbs and Hooten 2015).

The highest overall occupancy was found for the integrated model because ARUs and cameras each had at least one Sandhill Crane detection at a location that the other sensor type did not. Garland et al. (2020) found the same result (higher occupancy and detection probability for ARUs over cameras, but highest occupancy for the integrated data set) when comparing ARU and camera data for wolves (Canis lupus) in northern Alberta. They also posited that this was because of the difference in effective area surveyed. Our top integrated model included covariates at both spatial scales, indicating that both data sources contributed to improving understanding of crane habitat use, with cameras reflecting smaller-scale use and ARUs larger-scale use.

In ecological surveys, a species’ true abundance or occupancy across a study area can rarely be observed directly, only estimated from patterns derived from the data available. We were thus unable to evaluate which model provided the best estimate of occupancy; however, we compared the models to determine whether including a second dataset in an integrated model improved predictive performance over a model fit using only one dataset. In this case, K-fold cross-validation showed that integrating data sources did not improve predictive performance for ARU data, which is to say that the integrated model did not predict ARU data better than the ARU-only model did. In contrast, the integrated model did improve predictive performance for camera data (Table 2).

The pattern of the integrated map is very similar to the pattern of the ARU map. However, it does help tease apart patch- versus landscape-scale habitat characteristics, and the map shows specific low-quality habitat patches within high-quality landscape habitat (light points within darker areas). The Bayesian p-value of 0.00 for the camera fit in the integrated model is likely due to the difference in naive occupancy between ARU and camera detections; these goodness-of-fit metrics are still an active area of research in integrated models and the values shown here may not be reliable estimates of model fit (Doser et al. 2022).

Detection probability was lower in the integrated model than for models based on each sensor type alone. This is because although detection probability is modeled separately for each data source in an integrated model, it is conditional on the latent occurrence of both data sources combined, and each method detected cranes at at least one location that the other missed. This especially affected detection probability estimated from the camera data, which decreased from p = 0.08 to 0.03 when integrated with the ARU data. This could indicate that camera data have a higher probability of false negatives; however, as discussed above, detection probability for ARUs may also be biased high through detections of cranes whose territories were outside the buffer that we used to define the sampling location for the pair of sensors. We note that modeling detection probability jointly using both data sources could be a useful extension for these kinds of integrated models.

Despite the apparent redundancy of our integrated model compared to ARU data alone, both sampling methods have their own merits. Each sensor could be useful in different management applications, either of Sandhill Cranes or of other species such as Whooping Cranes. For example, ARUs could be used to detect larger-scale range expansions of Whooping Cranes back into their historical breeding range. Cameras, on the other hand, could help uncover more locally specific information on how cranes are using specific wetland microhabitats to inform wetland habitat management decisions such as installing water control devices. Future work could examine whether hierarchical habitat models based on camera data, either alone or in conjunction with ARU data, could make inferences about habitat selection of specific patches within a crane’s home range by including point-level habitat covariates (Johnson 1980).

Furthermore, cameras and ARUs have a range of uses that we did not explore here, which could be applied to many species. For example, cameras can be used to quantify behavior with respect to environmental stressors (Burton et al. 2022); they could also be used in a multi-state occupancy framework that separates breeding individuals from non-breeding (Nichols et al. 2007) or healthy individuals from unhealthy (Murray et al. 2021). Camera timelapse images have also been used to link vegetation phenology to Sandhill Crane occurrence (Sun et al. 2021). Autonomous recording units, in some cases, are able to discriminate individuals by unique acoustic cues, enabling more accurate density estimation (Larsen et al. 2022). Both methods can be used to monitor across multiple species and collect information on human disturbance (Buxton et al. 2018).

In this study, we focused on Sandhill Cranes. However, many other species have the potential to be studied using both methods. Several birds and mammals are seasonally vocal in Tuyeta outside of the time period for which we transcribed recordings, including moose, lynx, grouse, ptarmigans, and wolves. Targeted audio transcription during the peaks of vocalization for these species could be used to augment or compare with camera data. Autonomous recording units and cameras may also provide different types of inferences for such species. For example, ARUs could be used to identify the locations of Sharp-tailed Grouse (Tympanuchus phasianellus) leks, which are important to conserve for the long-term persistence of the species (Connelly et al. 2020). Cameras, however, might better inform year-round habitat use because Sharp-tailed Grouse are largely silent away from leks (Connelly et al. 2020). For very rare or endangered species, the additional detections obtained by both sampling methods could be of high value at limited additional costs, especially in remote regions where much of the cost is related to access (Mallory et al. 2018). For example, we obtained detections of the endangered Rusty Blackbird (Euphagus carolinus) on 10 cameras and 10 ARUs, representing 18 unique locations, so the addition of a second sensor was non-trivial. It remains to be seen whether, and how much, combining ARU and camera trap data could improve inferences for a larger number of species that differ from Sandhill Cranes in vocalization characteristics and other attributes.

Conclusion

Habitat models built using ARU data allowed for inference about patterns of Sandhill Crane occupancy at a larger scale than habitat models made with camera trap data, although an integrated model reflected both small and large spatial scales. Autonomous recording units models had higher detection probability than camera trap models, likely due to the loud vocalizations of Sandhill Cranes. Accordingly, the integrated model did not improve predictive performance over ARU data alone, and its predictions of occupancy probability across the study area were similar to the ARU model. Although our study does not provide evidence that an integrated model using data from both sampling methods improves inferences for Sandhill Crane habitat use, data from each method provided complementary information about Sandhill Crane ecology. We recommend that ARUs be used to achieve the highest detection probability for such highly vocal species, or when examining landscape-scale predictors of occupancy. We recommend that camera traps be used when examining factors affecting habitat-patch scale occupancy. Overall, both ARUs and camera traps are cost-effective and non-invasive methods for assessing species distributions and habitat selection. We anticipate that combining these methods can increase the breadth of species covered in monitoring programs, while yielding more nuanced insights on the ecology of species frequently detected by both sight and sound.

LITERATURE CITED

Allen, R. P. 1952. The Whooping Crane. Research report no. 3 of the National Audubon Society. National Audubon Society, New York, New York, USA. https://savingcranes.org/wp-content/uploads/2024/11/the_whooping_crane_porter_allen.pdf

Ausband, D. E., L. N. Rich, E. M. Glenn, M. S. Mitchell, P. Zager, D. A. W. Miller, L. P. Waits, B. B. Ackerman, and C. M. Mack. 2014. Monitoring gray wolf populations using multiple survey methods. Journal of Wildlife Management 78:335–346. https://doi.org/10.1002/jwmg.654

Baker, B. W., B. S. Cade, W. L. Mangus, and J. L. McMillen. 1995. Spatial analysis of Sandhill Crane nesting habitat. Journal of Wildlife Management 59:752–758. https://doi.org/10.2307/3801952

Barber-Meyer, S. M., V. Palacios, B. Marti-Domken, and L. J. Schmidt. 2020. Testing a new passive acoustic recording unit to monitor wolves. Wildlife Society Bulletin 44:590–598. https://doi.org/10.1002/wsb.1117

Becker, M., D. J. Huggard, M. Dickie, C. Warbington, J. Schieck, E. Herdman, R. Serrouya, and S. Boutin. 2022. Applying and testing a novel method to estimate animal density from motion‐triggered cameras. Ecosphere 13:E4005. https://doi.org/10.1002/ecs2.4005

Beirne, C., C. Sun, E. R. Tattersall, J. M. Burgar, J. T. Fisher, and A. C. Burton. 2021. Multispecies modelling reveals potential for habitat restoration to re‐establish boreal vertebrate community dynamics. Journal of Applied Ecology 58:2821–2832. https://doi.org/10.1111/1365-2664.14020

Bleich, V. C., R. T. Bowyer, A. M. Pauli, M. C. Nicholson, and R. W. Anthes. 1994. Mountain sheep Ovis canadensis and helicopter surveys: ramifications for the conservation of large mammals. Biological Conservation 70:1–7. https://doi.org/10.1016/0006-3207(94)90292-5

Bobrowski, M., J. Weidinger, N. Schwab, and U. Schickhoff. 2021. Searching for ecology in species distribution models in the Himalayas. Ecological Modelling 458:109693. https://doi.org/10.1016/j.ecolmodel.2021.109693

Brambilla, A., and F. Brivio. 2018. Assessing the effects of helicopter disturbance in a mountain ungulate on different time scales. Mammalian Biology 90:30–37. https://doi.org/10.1016/j.mambio.2018.02.001

Brooks, S. P., and A. Gelman. 1998. General methods for monitoring convergence of iterative simulations. Journal of Computational and Graphical Statistics 7:434–455. https://doi.org/10.1080/10618600.1998.10474787

Buckland, S. T., D. R. Anderson, K. P. Burnham, J. L. Laake, D. L. Borchers, and L. Thomas. 2001. Introduction to distance sampling: estimating abundance of biological populations. Oxford University Press, Oxford, UK. https://doi.org/10.1093/oso/9780198506492.001.0001

Burton, A. C., C. Beirne, C. Sun, A. Granados, M. Procko, C. Chen, M. Fennell, A. Constantinou, C. Colton, K. Tjaden‐McClement, J. T. Fisher, and J. Burgar. 2022. Behavioral “bycatch” from camera trap surveys yields insights on prey responses to human‐mediated predation risk. Ecology and Evolution 12:e9108. https://doi.org/10.1002/ece3.9108

Burton, A. C., D. Huggard, E. Bayne, J. Schieck, P. Sólymos, T. Muhly, D. Farr, and S. Boutin. 2014. A framework for adaptive monitoring of the cumulative effects of human footprint on biodiversity. Environmental Monitoring and Assessment 186:3605–3617. https://doi.org/10.1007/s10661-014-3643-7

Burton, A. C., E. Neilson, D. Moreira, A. Ladle, R. Steenweg, J. T. Fisher, E. Bayne, and S. Boutin. 2015. Wildlife camera trapping: a review and recommendations for linking surveys to ecological processes. Journal of Applied Ecology 52:675–685. https://doi.org/10.1111/1365-2664.12432

Buxton, R. T., R. Galvan, M. F. McKenna, C. L. White, and V. Seher. 2017. Visitor noise at a nesting colony alters the behavior of a coastal seabird. Marine Ecology Progress Series 570:233–246. https://doi.org/10.3354/meps12073

Buxton, R. T., P. E. Lendrum, K. R. Crooks, and G. Wittemyer. 2018. Pairing camera traps and acoustic recorders to monitor the ecological impact of human disturbance. Global Ecology and Conservation 16:e00493. https://doi.org/10.1016/j.gecco.2018.e00493

Canadian Wildlife Service. 2007. Ts’ude niline Tu’eyeta: ecological assessment of the Ts’ude niline Tu’eyeta candidate protected area. Canadian Wildlife Service, Environment Canada, Yellowknife, Northwest Territories, Canada. https://nwtdiscoveryportal.enr.gov.nt.ca/geoportaldocuments/document-2007-ramparts-eafull.pdf

Connelly, J. W., M. W. Gratson, and K. P. Reese. 2020. Sharp-tailed Grouse Tympanuchus phasianellus. In S. M. Billerman, B. K. Keeney, P. G. Rodewald, and T. S. Schulenberg, editors. Birds of the world. Cornell Lab of Ornithology, Lansing, New York, USA. https://doi.org/10.2173/bow.shtgro.01

Conway, C. J., and J. P. Gibbs. 2005. Effectiveness of call-broadcast surveys for monitoring marsh birds. Auk 122:26–35. https://doi.org/10.1093/auk/122.1.26

Ditmer, M. A., J. B. Vincent, L. K. Werden, J. C. Tanner, T. G. Laske, P. A. Iaizzo, D. L. Garshelis, and J. R. Fieberg. 2015. Bears show a physiological but limited behavioral response to unmanned aerial vehicles. Current Biology 25:2278–2283. https://doi.org/10.1016/j.cub.2015.07.024

Dobbins, M., R. Sollmann, S. Menke, A. A. Zambrano, and E. Broadbent. 2020. An integrated approach to measure hunting intensity and assess its impacts on mammal populations. Journal of Applied Ecology 57:2100–2111. https://doi.org/10.1111/1365-2664.13750

Dormann, C. F., J. Elith, S. Bacher, C. Buchmann, G. Carl, G. Carré, J. R. García Marquéz, B. Gruber, B. Lafourcade, P. J. Leitão, T. Münkemüller, C. McClean, P. E. Osborne, B. Reineking, B. Schröder, A. K. Skidmore, D. Zurell, and S. Lautenbach. 2013. Collinearity: a review of methods to deal with it and a simulation study evaluating their performance. Ecography 36:27–46. https://doi.org/10.1111/j.1600-0587.2012.07348.x

Doser, J. W., A. O. Finley, M. Kéry, and E. F. Zipkin. 2022. spOccupancy: an R package for single‐species, multi‐species, and integrated spatial occupancy models. Methods in Ecology and Evolution 13:1670–1678. https://doi.org/10.1111/2041-210X.13897

Drake, A., D. R. de Zwaan, T. A. Altamirano, S. Wilson, K. Hick, C. Bravo, J. T. Ibarra, and K. Martin. 2021. Combining point counts and autonomous recording units improves avian survey efficacy across elevational gradients on two continents. Ecology and Evolution 11:8654–8682. https://doi.org/10.1002/ece3.7678

Drewien, R. C. 1973. Ecology of the Rocky Mountain Greater Sandhill Cranes. Dissertation. University of Idaho, Moscow, Idaho, USA.

Ducks Unlimited Canada. 2021. Ramparts enhanced wetland classification user’s guide. Ducks Unlimited Canada, Edmonton, Alberta, Canada.

Ecological Stratification Working Group. 1995. A national ecological framework for Canada. Agriculture and Agri-Food Canada, Research Branch, Centre for Land and Biological Resources Research and Environment Canada, State of the Environment Directorate, Ecozone Analysis Branch, Ottawa, Ontario, Canada. https://sis.agr.gc.ca/cansis/publications/manuals/1996/A42-65-1996-national-ecological-framework.pdf

Environment and Natural Resources. 2023. Biodiversity monitoring program. Government of the Northwest Territories, Yellowknife, Northwest Territories, Canada. https://www.enr.gov.nt.ca/en/services/biodiversity/biodiversity-monitoring-program

Fitch, W. T. 1999. Acoustic exaggeration of size in birds via tracheal elongation: comparative and theoretical analyses. Journal of Zoology 248:31–48. https://doi.org/10.1111/j.1469-7998.1999.tb01020.x

Fleming, P. J. S., and J. P. Tracey. 2008. Some human, aircraft and animal factors affecting aerial surveys: how to enumerate animals from the air. Wildlife Research 35:258–267. https://doi.org/10.1071/WR07081

Fontúrbel, F. E., G. B. Rodríguez-Gómez, N. Fernández, B. García, J. I. Orellana, and G. J. Castaño-Villa. 2020. Sampling understory birds in different habitat types using point counts and camera traps. Ecological Indicators 119:106863. https://doi.org/10.1016/j.ecolind.2020.106863

Francomano, D., A. N. Raya Rey, B. L. Gottesman, and B. C. Pijanowski. 2024. Acoustic recording complements camera traps for monitoring sensitive penguin populations. Ibis 166:38–54. https://doi.org/10.1111/ibi.13235

Garland, L., A. Crosby, R. Hedley, S. Boutin, and E. Bayne. 2020. Acoustic vs. photographic monitoring of gray wolves (Canis lupus): a methodological comparison of two passive monitoring techniques. Canadian Journal of Zoology 98:219–228. https://doi.org/10.1139/cjz-2019-0081

Gerber B. D., J. F. Dwyer, S. A. Nesbitt, R. C. Drewien, C. D. Littlefield, T. C. Tacha, and P. A. Vohs. 2020. Sandhill Crane Antigone canadensis. In S. M. Billerman, B. K. Keeney, P. G. Rodewald, and T. S. Schulenberg, editors. Birds of the world. Cornell Lab of Ornithology, Lansing, New York, USA. https://doi.org/10.2173/bow.sancra.01

Gorresen, P. M., P. M. Cryan, K. Montoya‐Aiona, and F. J. Bonaccorso. 2017. Do you hear what I see? Vocalization relative to visual detection rates of Hawaiian hoary bats (Lasiurus cinereus semotus). Ecology and Evolution 7:6669–6679. https://doi.org/10.1002/ece3.3196

Government of Canada. 1994. Migratory birds convention act, 22 SC. 1994. Government of Canada, Ottawa, Ontario, Canada. https://laws-lois.justice.gc.ca/eng/acts/m-7.01/

Government of Canada. 2002. Species at Risk Act, 29 SC. 2002. Government of Canada, Ottawa, Ontario, Canada. https://laws.justice.gc.ca/eng/acts/s-15.3/

Hanson, J. O., J. L. McCune, I. Chadès, C. A. Proctor, E. J. Hudgins, and J. R. Bennett. 2023. Optimizing ecological surveys for conservation. Journal of Applied Ecology 60:41–51. https://doi.org/10.1111/1365-2664.14309

Hobbs, N. T., and M. B. Hooten. 2015. Bayesian models: a statistical primer for ecologists. Princeton University Press, Princeton, New Jersey, USA. https://doi.org/10.23943/princeton/9780691159287.003.0005

Hooten, M. B., and N. T. Hobbs. 2015. A guide to Bayesian model selection for ecologists. Ecological Monographs 85:3–28. https://doi.org/10.1890/14-0661.1

Isaac, N. J. B., M. A. Jarzyna, P. Keil, L. I. Dambly, P. H. Boersch-Supan, E. Browning, S. N. Freeman, N. Golding, G. Guillera-Arroita, P. A. Henrys, S. Jarvis, J. Lahoz-Monfort, J. Pagel, O. L. Pescott, R. Schmucki, E. G. Simmonds, and R. B. O–Hara. 2020. Data integration for large-scale models of species distributions. Trends in Ecology and Evolution 35:56–67. https://doi.org/10.1016/j.tree.2019.08.006

International Union for the Conservation of Nature (IUCN). 2022. The IUCN Red List of threatened species. Version 2022-1. IUCN, Gland, Switzerland. https://www.iucnredlist.org

Johnson, D. H. 1980. The comparison of usage and availability measurements for evaluating resource preference. Ecology 61:65–71. https://doi.org/10.2307/1937156

Kays, R., M. H. Snider, G. Hess, M. V. Cove, A. Jensen, H. Shamon, W. J. McShea, B. Rooney, M. L. Allen, C. E. Pekins, C. C. Wilmers, M. E. Pendergast, A. M. Green, J. Suraci, M. S. Leslie, S. Nasrallah, D. Farkas, M. Jordan, M. Grigione, M. C. LaScaleia, M. L. Davis, C. Hansen, J. Millspaugh, J. S. Lewis, M. Havrda, R. Long, K. R. Remine, K. J. Jaspers, D. J. R. Lafferty, T. Hubbard, C. E. Studds, E. L. Barthelmess, K. Andy, A. Romero, B. J. O’Neill, M. T. R. Hawkins, J. V. Lombardi, M. Sergeyev, M. C. Fisher-Reid, M. S. Rentz, C. Nagy, J. M. Davenport, C. C. Rega-Brodsky, C. L. Appel, D. B. Lesmeister, S. T. Giery, C. A. Whittier, J. M. Alston, C. Sutherland, C. Rota, T. Murphy, T. E. Lee, Jr., A. Mortelliti, D. L. Bergman, J. A. Compton, B. D. Gerber, J. Burr, K. Rezendes, B. A. DeGregorio, N. H. Wehr, J. F. Benson, M. T. O’Mara, D. S. Jachowski, M. Gray, D. E. Beyer, Jr., J. L. Belant, R. V. Horan, III, R. C. Lonsinger, K. M. Kuhn, S. C. M. Hasstedt, M. Zimova, S. M. Moore, D. J. Herrera, S. Fritts, A. J. Edelman, E. A. Flaherty, T. R. Petroelje, S. A. Neiswenter, D. R. Risch, F. Iannarilli, M. van der Merwe, S. P. Maher, Z. J. Farris, S. L. Webb, D. S. Mason, M. A. Lashley, A. M. Wilson, J. P. Vanek, S. R. Wehr, L. M. Conner, J. C. Beasley, H. L. Bontrager, C. Baruzzi, S. N. Ellis-Felege, M. D. Proctor, J. Schipper, K. C. B. Weiss, A. K. Darracq, E. G. Barr, P. D. Alexander, Ç. H. Şekercioğlu, D. A. Bogan, C. M. Schalk, J. E. Fantle-Lepczyk, C. A. Lepczyk, S. LaPoint, L. S. Whipple, H. I. Rowe, K. Mullen, T. Bird, A. Zorn, L. Brandt, R. G. Lathrop, C. McCain, A. P. Crupi, J. Clark, and A. Parsons. 2024. Climate, food and humans predict communities of mammals in the United States. Diversity and Distributions 30:e13900. https://doi.org/10.1111/ddi.13900

Kéry, M., and J. A. Royle. 2016. Applied hierarchical modeling in ecology. Volume 1: prelude and static models. Elsevier Science, Amsterdam, The Netherlands.

Kleist, N. J., R. T. Buxton, P. E. Lendrum, C. Linares, K. R. Crooks, and G. Wittemyer. 2020. Noise and landscape features influence habitat use of mammalian herbivores in a natural gas field. Journal of Animal Ecology 90:875–885. https://doi.org/10.1111/1365-2656.13416

Kruse, K. L., D. P. Collins, C. M. Conring, B. A. Grisham, W. C. Conway, and J. M. Knetter. 2017. Summer habitat selection of the Lower Colorado River Valley population of Greater Sandhill Cranes. Journal of Fish and Wildlife Management 8:436–448. https://doi.org/10.3996/042017-JFWM-037

Larsen, H. L., C. Pertoldi, N. Madsen, E. Randi, A. V. Stronen, H. Root-Gutteridge, and S. Pagh. 2022. Bioacoustic detection of wolves: identifying subspecies and individuals by howls. Animals 12:631. https://doi.org/10.3390/ani12050631

Lunn, D., C. Jackson, N. Best, A. Thomas, and D. Spiegelhalter. 2013. The bugs book: a practical introduction to Bayesian analysis. https://doi.org/10.1201/b13613

MacKenzie, D. I., J. D. Nichols, J. A. Royle, K. H. Pollock, L. L. Bailey, and J. E. Hines. 2017. Occupancy estimation and modeling: inferring patterns and dynamics of species occurrence. Second edition. Academic, Cambridge, Massachusetts, USA. https://doi.org/10.1016/C2012-0-01164-7

Mallory, M. L., H. G. Gilchrist, M. Janssen, H. L. Major, F. Merkel, J. F. Provencher, and H. Strøm. 2018. Financial costs of conducting science in the Arctic: examples from seabird research. Arctic Science 4:624–633. https://doi.org/10.1139/as-2017-0019

Matsuoka, S. M., E. M. Bayne, P. Sólymos, P. C. Fontaine, S. G. Cumming, F. K. A. Schmiegelow, and S. J. Song. 2012. Using binomial distance-sampling models to estimate the effective detection radius of point-count surveys across boreal Canada. Auk 129:268–282. https://doi.org/10.1525/auk.2012.11190

Mayor, S. J., D. C. Schneider, J. A. Schaefer, and S. P. Mahoney. 2009. Habitat selection at multiple scales. Écoscience 16:238–247. https://doi.org/10.2980/16-2-3238

Meek, P. D., G. A. Ballard, and G. Falzon. 2016. The higher you go the less you will know: placing camera traps high to avoid theft will affect detection. Remote Sensing in Ecology and Conservation 2:204–211. https://doi.org/10.1002/rse2.28

Melvin, S. M., W. J. D. Stephen, and S. A. Temple. 1990. Population estimates, nesting biology, and habitat preferences of Interlake, Manitoba, Sandhill Cranes, Grus canadensis. Canadian Field-Naturalist 104:354–361. https://doi.org/10.5962/p.356395

Microsoft. 2020. AI for Earth camera trap image processing API (4.1). Microsoft, Redmond, Washington, USA. https://github.com/microsoft/CameraTraps/blob/main/megadetector.md

Miller, D. A. W., K. Pacifici, J. S. Sanderlin, and B. J. Reich. 2019. The recent past and promising future for data integration methods to estimate species’ distributions. Methods in Ecology and Evolution 10:22–37. https://doi.org/10.1111/2041-210X.13110

Murray, M. H., M. Fidino, E. W. Lehrer, J. L. Simonis, and S. B. Magle. 2021. A multi‐state occupancy model to non‐invasively monitor visible signs of wildlife health with camera traps that accounts for image quality. Journal of Animal Ecology 90:1973–1984. https://doi.org/10.1111/1365-2656.13515

Neilson, E. W., T. Avgar, A. C. Burton, K. Broadley, and S. Boutin. 2018. Animal movement affects interpretation of occupancy models from camera-trap surveys of unmarked animals. Ecosphere 9:e02092. https://doi.org/10.1002/ecs2.2092

Nichols, J. D., J. E. Hines, D. I. Mackenzie, M. E. Seamans, and R. J. Gutiérrez. 2007. Occupancy estimation and modeling with multiple states and uncertainty. Ecology 88:1395–1400. https://doi.org/10.1890/06-1474

Nichols, J. D., and B. K. Williams. 2006. Monitoring for conservation. Trends in Ecology and Evolution 21:668–673. https://doi.org/10.1016/j.tree.2006.08.007

Northrup, J. M., C. R. Anderson, Jr., and G. Wittemyer. 2014. Effects of helicopter capture and handling on movement behavior of mule deer: capture effects on mule deer. Journal of Wildlife Management 78:731–738. https://doi.org/10.1002/jwmg.705

Palacios, V., S. M. Barber-Meyer, B. Martí-Domken, and L. J. Schmidt. 2022. Assessing spontaneous howling rates in captive wolves using automatic passive recorders. Bioacoustics 31:562–574. https://doi.org/10.1080/09524622.2021.2006083

Rayment, W., T. Webster, T. Brough, T. Jowett, and S. Dawson. 2018. Seen or heard? A comparison of visual and acoustic autonomous monitoring methods for investigating temporal variation in occurrence of southern right whales. Marine Biology 165:12. https://doi.org/10.1007/s00227-017-3264-0

R Core Team. 2022. R: a language and environment for statistical computing. R Foundation for Statistical Computing, Vienna, Austria. https://www.R-project.org

Reed, J. R. 1988. Arctic adaptations in the breeding biology of Sandhill Cranes, Grus canadensis, on Banks Island, Northwest Territories. Canadian Field-Naturalist 102:643–648. https://doi.org/10.5962/p.356643

Rhodes, J. R., C. A. McAlpine, A. F. Zuur, G. M. Smith, and E. N. Ieno. 2009. GLMM applied on the spatial distribution of koalas in a fragmented landscape. Pages 469–492 in A. F. Zuur, E. N. Ieno, N. Walker, A. A. Saveliev, and G. M. Smith, editors. Mixed effects models and extensions in ecology with R. Springer, New York, New York, USA. https://doi.org/10.1007/978-0-387-87458-6_21

Saunders, S. P., K. A. L. Hall, N. Hill, and N. L. Michel. 2019. Multiscale effects of wetland availability and matrix composition on wetland breeding birds in Minnesota, USA. Condor 121:duz024. https://doi.org/10.1093/condor/duz024

Schlossberg, S., M. J. Chase, and C. R. Griffin. 2016. Testing the accuracy of aerial surveys for large mammals: an experiment with African savanna elephants (Loxodonta africana). PLoS ONE 11:0164904. https://doi.org/10.1371/journal.pone.0164904

Shonfield, J., and E. M. Bayne. 2017. Autonomous recording units in avian ecological research: current use and future applications. Avian Conservation and Ecology 12(1):14. https://doi.org/10.5751/ACE-00974-120114

Smith, K. B., C. E. Smith, S. F. Forest, and A. J. Richard. 2007. Field guide to the wetlands of the boreal plains ecozone of Canada. Ducks Unlimited Canada, Western Boreal Office, Edmonton, Alberta, Canada. https://www.ducks.ca/assets/2015/09/fieldguide-wetlands-boreal-plains.pdf

Sólymos, P., S. M. Matsuoka, E. M. Bayne, S. R. Lele, P. Fontaine, S. G. Cumming, D. Stralberg, F. K. A. Schmiegelow, and S. J. Song. 2013. Calibrating indices of avian density from non-standardized survey data: making the most of a messy situation. Methods in Ecology and Evolution 4:1047–1058. https://doi.org/10.1111/2041-210X.12106

Soulsbury, C. D., H. E. Gray, L. M. Smith, V. Braithwaite, S. C. Cotter, R. W. Elwood, A. Wilkinson, and L. M. Collins. 2020. The welfare and ethics of research involving wild animals: a primer. Methods in Ecology and Evolution 11:1164–1181. https://doi.org/10.1111/2041-210X.13435

Spak, S. 2005. The position of Indigenous knowledge in Canadian co-management organizations. Anthropologica 47:233–246. https://cas-sca.journals.uvic.ca/index.php/anthropologica/article/view/2385/2171

Stewart, F. E. C., J. P. Volpe, and J. T. Fisher. 2019. The debate about bait: a red herring in wildlife research. Journal of Wildlife Management 83:985–992. https://doi.org/10.1002/jwmg.21657

Stiffler, L. L., J. T. Anderson, and T. E. Katzner. 2018. Occupancy modeling of autonomously recorded vocalizations to predict distribution of rallids in tidal wetlands. Wetlands 38:605–612. https://doi.org/10.1007/s13157-018-1003-z

Sun, C., C. Beirne, J. M. Burgar, T. Howey, J. T. Fisher, and A. C. Burton. 2021. Simultaneous monitoring of vegetation dynamics and wildlife activity with camera traps to assess habitat change. Remote Sensing in Ecology and Conservation 7:666–684. https://doi.org/10.1002/rse2.222

Taylor, M. F. J., K. F. Suckling, and J. J. Rachlinski. 2005. The effectiveness of the Endangered Species Act: a quantitative analysis. BioScience 55:360–367. https://doi.org/10.1641/0006-3568(2005)055[0360:TEOTES]2.0.CO;2

Toews, M., F. Juanes, and A. C. Burton. 2017. Mammal responses to human footprint vary with spatial extent but not with spatial grain. Ecosphere 8:e01735. https://doi.org/10.1002/ecs2.1735

U.S. National Vegetation Classification (USNVC). 2016. United States National Vegetation classification database, V2.0. Federal Geographic Data Committee, Reston, Virginia, USA. http:/usnvc.org

Van Wilgenburg, S. L., C. L. Mahon, G. Campbell, L. McLeod, M. Campbell, D. Evans, W. Easton, C. M. Francis, S. Haché, C. S. Machtans, C. Mader, R. F. Pankratz, R. Russell, A. C. Smith, P. Thomas, J. D. Toms, and J. A. Tremblay. 2020. A cost efficient spatially balanced hierarchical sampling design for monitoring boreal birds incorporating access costs and habitat stratification. PLoS ONE 15:0234494. https://doi.org/10.1371/journal.pone.0234494

Van Wilgenburg, S. L., P. Sólymos, K. J. Kardynal, and M. D. Frey. 2017. Paired sampling standardizes point count data from humans and acoustic recorders. Avian Conservation and Ecology 12(1):13. https://doi.org/10.5751/ACE-00975-120113

Vaughan, P. M., J. C. Buettel, and B. W. Brook. 2022. Investigating avian behaviour using opportunistic camera-trap imagery reveals an untapped data source. Ornithological Science 21:3–12. https://doi.org/10.2326/osj.21.3

Vélez, J., W. McShea, B. Pukazhenthi, P. Stevenson, and J. Fieberg. 2024. Implications of the scale of detection for inferring co‐occurrence patterns from paired camera traps and acoustic recorders. Conservation Biology 38:e14218. https://doi.org/10.1111/cobi.14218

Walkinshaw, L. H. 1973. Cranes of the world. Winchester, New York, New York, USA .

Wevers, J., N. Beenaerts, J. Casaer, F. Zimmermann, T. Artois, and J. Fattebert. 2021. Modelling species distribution from camera trap by‐catch using a scale‐optimized occupancy approach. Remote Sensing in Ecology and Conservation 7:534–549. https://doi.org/10.1002/rse2.207

Wheatley, M., and C. Johnson. 2009. Factors limiting our understanding of ecological scale. Ecological Complexity 6:150–159. https://doi.org/10.1016/j.ecocom.2008.10.011

White, G. 2020. Indigenous empowerment through co-management: land claims boards, wildlife management, and environmental regulation. UBC Press, Vancouver, British Columbia, Canada. https://doi.org/10.59962/9780774863049

Wiens, J. A. 1989. Spatial scaling in ecology. Functional Ecology 3:385–397. https://doi.org/10.2307/2389612

Wong, C., K. Ballegooyen, L. Ignace, M. J. Johnson, and H. Swanson. 2020. Towards reconciliation: 10 Calls to Action to natural scientists working in Canada. FACETS 5:769–783. https://doi.org/10.1139/facets-2020-0005