Skip to main content
NCBI home page
Springer logoLink to Springer
. 2025 Sep 20;197(10):1129. doi: 10.1007/s10661-025-14581-7

Camera trapping for density estimation: comparing the TIFC model to aerial surveys for multiple ungulate populations

Jennifer M Foca 1,✉, Darcy R Visscher 1,2, Marcus Becker 3, Mark S Boyce 1
PMCID: PMC12450223  PMID: 40974373

Abstract

Population density estimates are invaluable to wildlife managers, but difficult to attain. Several methods have been developed to estimate density using camera traps, many of which require further testing. The time-in-front-of-the-camera (TIFC) approach allows for density estimation when “unmarked” individuals are monitored using camera traps. We applied the TIFC model to populations of bison (Bison bison, B. bison athabascae), elk (Cervus elaphus canadensis), and moose (Alces alces) in Elk Island National Park (EINP) and to populations of elk and moose in Cooking Lake – Blackfoot Provincial Recreation Area (BPRA). EINP and BPRA are fully fenced natural areas in the Beaverhills Region of central Alberta, Canada. Our objectives were to (i) use the TIFC model to estimate ungulate densities in EINP and BPRA, and (ii) compare the performance of TIFC density estimates against aerial ungulate survey estimates. Camera trap data were collected from 43 cameras in EINP between December 2016 and October 2020, and 23 cameras in BPRA from April 2019 to August 2020. Annual densities were estimated in EINP north and south (2017–2019) and in BPRA (2019). Moose density estimates had the lowest discrepancy between approaches. Bison TIFC density estimates were lower than AUS densities, and elk TIFC density estimates were higher than AUS densities. In addition to the density estimates evaluated for the three focal species, the TIFC approach also was applied to white-tailed deer (Odocoileus virginianus) and mule deer (O. hemionus) in EINP and BPRA, in the absence of aerial survey data. We conclude that the TIFC model and AUS were complementary, with pros and cons of the two approaches varying based on focal species ecology. Careful consideration is required for several factors related to camera study design for TIFC density estimation that can affect the accuracy and precision of estimates.

Keywords: Camera traps, Density estimation, Population monitoring, Time in front of the camera, Ungulates, Wildlife management

Introduction

Estimating density or abundance of wild populations is a common goal of wildlife managers (Williams et al., 2002). There also is a need for monitoring approaches that can be applied to multiple species at the same time. Aerial surveys have been used to estimate ungulate density or abundance; however, aerial surveys have limitations including visibility bias (Caughley, 1974; Caughley et al., 1976). Camera traps are increasingly used by wildlife managers, with many promising applications for density estimation (Burton et al., 2015).

In Canada, aerial ungulate surveys (AUS) are the predominant method for estimating ungulate densities (Boyce et al., 2012; Gasaway et al., 1986; Habib et al., 2012; McIntosh et al., 2009; Steinhorst & Samuel, 1989). Advantages of aerial surveys include excellent spatial coverage and a relatively low time commitment. Drawbacks include the cost of aircraft time, which often results in poor temporal coverage, with aerial surveys providing only a snapshot in time. Aerial surveys are usually constrained to certain times of year, specific flying conditions, and have limitations like minimum snow cover (Lynch & Shumaker, 1995; Allen et al., 2008). For aerial surveys, sightability varies greatly based on species, group size, vegetation cover, animal behavior, and flying conditions (Anderson & Lindzey, 1996; Graham & Bell, 1989; Samuel et al., 1987). Aerial surveys have limited ability to collect information on sex ratios for some ungulate species, data that can be used in ungulate population management (Bender, 2006). Lastly, aerial surveys are much more dangerous for researchers than camera trap surveys (Sasse, 2003).

In contrast, camera traps can yield excellent temporal coverage relative to the amount of field effort required because data can be collected remotely throughout the year. Camera-trap surveys have great potential for collecting detailed demographic data (i.e., sex ratios, age class), and are better for collecting data on species in closed cover areas. Some challenges related to the use of camera trap surveys include that spatial coverage is often more limited for camera traps than aerial surveys because researchers are generally limited by the number of cameras available to deploy. Additionally, camera trap surveys can have issues with imperfect detection, unequal detection probabilities, and other sources of bias (Burton et al., 2015). Overall, camera traps have potential for estimating densities for a broader range of species and time frames, though many camera trap methods still require assessment.

Most studies on density estimation using camera traps have focused on techniques for “marked” populations (Burton et al., 2015), such as for species where individuals have uniquely identifiable coat patterns. Camera trap density methods for marked populations include conventional capture-recapture (CR; Karanth & Nichols, 1998) and spatial capture-recapture (SCR; Efford, 2004; Borchers & Efford, 2008Royle et al., 2014). Approaches for estimating population densities for unmarked species have expanded camera trap applicability and include the Random Encounter Model (REM; Rowcliffe et al., 2008), spatial-count models (SCM; Chandler & Royle, 2013), space-to-event model (STE; Moeller et al., 2018), Random Encounter and Staying Time model (REST; Nakashima et al., 2018), and time-in-front-of-the-camera (TIFC) model (Becker et al., 2022; Warbington & Boyce, 2020). REM, REST, and TIFC have similarities, including that each requires estimates of the area sampled in the camera field of view. REM uses detection rates and movement rates to estimate densities, whereas REST and TIFC use the cumulative animal staying time in the camera trap field-of-view. REST was intended for remote camera video data, whereas TIFC has been developed for photo data, making TIFC more widely applicable. Unlike the REST approach, the TIFC approach does not require parameterization of encounter rates and staying times and instead uses the cumulative staying time for any member of a given species during the monitoring period (Becker et al., 2022; Warbington & Boyce, 2020). TIFC has no requirements for animal movement rates, average group size, or home range size (Becker et al., 2022; Foster & Harmsen, 2012; Nakashima et al., 2018; Warbington & Boyce, 2020), making it an attractive and accessible approach for researchers and wildlife managers.

We compared population density estimates from camera traps and aerial surveys for multiple closed populations of ungulates in two fully fenced natural areas. The objectives of our study were to (i) estimate ungulate densities using TIFC (Becker et al., 2022; Warbington & Boyce, 2020) and (ii) compare the performance of TIFC density estimates against aerial ungulate survey estimates, which are more commonly used in wildlife management. The TIFC method was selected based on its potential applicability for wildlife researchers paired with accessibility because additional data sources (e.g., GPS collars for movement rates, home range sizes) are not required to apply the TIFC method. Our study took place in Elk Island National Park (EINP) and Cooking Lake—Blackfoot Provincial Recreation Area (BPRA), two fully fenced natural areas located in the Beaver Hills UNESCO Biosphere Reserve in Alberta, Canada, with five unmarked ungulates: elk (Cervus elaphus canadensis), moose (Alces alces), mule deer (Odocoileus hemionus), white-tailed deer (O. virginianus), and bison (Bison bison, B. bison athabascae), with bison being present only in EINP. This allowed us to apply TIFC to multiple closed populations of ungulates across multiple years (2017–2019 for EINP, 2019 for BPRA). Aerial surveys have been conducted intensively in EINP each winter attempting a total count, and during 2019–2020, we extended the AUS to BPRA as well. We were able to calculate TIFC density estimates for all five ungulate species, but there are no reliable aerial survey estimates for white-tailed deer and mule deer due to sightability limitations. Therefore, we selected bison, elk, and moose as the three focal species to compare TIFC estimates to densities observed during aerial surveys. In this study, we provide the first comparison between aerial surveys and the TIFC method in a multispecies monitoring context.

Materials and methods

Study areas

Elk Island National Park (EINP) and Cooking Lake–Blackfoot Provincial Recreation Area (BPRA) are adjacent protected areas located in the Beaver Hills UNESCO Biosphere Reserve in central Alberta, Canada on Treaty 6 territory (Foca & Boyce, 2022). These natural areas are representative of the Southern Boreal Plains and Plateaux Natural Region (Parks Canada, 2010), which is characterized by glacial moraine topography with a mix of deciduous forests, lakes, wetlands, and grasslands. Forested areas in these natural areas are largely deciduous, dominated by trembling aspen (Populus tremuloides), with occasional clusters of conifers.

The Yellowhead Hwy, AB 16, divides EINP into separate northern and southern areas (EINP-N and EINP-S respectively; Fig. 1). BPRA shares its northern border with EINP-S. Each of these three areas is enclosed by a 2.2 m high fence that restricts wildlife movement in and out of each area. We therefore considered these areas as three independent study areas: EINP-N (134 km2), EINP-S (60 km2), and BPRA (97 km2). Vegetation cover is similar in all three study areas; however, in BPRA, approximately 50% of the landscape has been converted to grazing pastures. BPRA is a multiuse area with seasonal cattle grazing, Indigenous and licensed hunting, maintained gas wells, and a variety of opportunities for non-motorized recreation. Recreation opportunities are more limited in EINP, with hiking and biking being the most common types of recreation. Human use and human trail density are higher in EINP-N and BPRA compared to EINP-S.

Fig. 1.

Fig. 1

Open in a new tab

Camera trap survey across three study areas in central Alberta, Canada. Cameras were deployed on a 2× 2km2 systematic grid, shown in black. The three adjacent study areas include Elk Island National Park—north (EINP-N; 134km2), Elk Island National Park—south (EINP-S; 60km2), and Cooking Lake—Blackfoot Provincial Recreation Area (BPRA; 97km2). EINP-N and EINP-S are separated by Highway 16, shown in red. Each of the three study areas is enclosed by a 2.2-m high fence, restricting ungulate movement between areas

Elk, moose, mule deer, and white-tailed deer are present in all three study areas. Plains bison (Bison bison bison) are present only in EINP-N, while wood bison (B. bison athabascae) are present in EINP-S, keeping the populations genetically distinct. Predation and dispersal are limited in all three sites by fenced perimeters. Predators currently include transient black bears (Ursus americanus), resident coyotes (Canis latrans), and occasional wolves (C. lupus) and cougars (Puma concolor); but none is thought to have a limiting effect on ungulate populations. Active ungulate management in EINP is necessary to prevent populations from becoming overabundant (Parks Canada, 2017). Current ungulate management includes bison removals (i.e., translocations, culls), alternating between EINP-N and EINP-S in winter each year. Elk translocations are no longer allowed due to the risk of spreading chronic wasting disease (Parks Canada, 2017). Unlike BPRA, EINP does not allow hunting as part of its ungulate management. In addition, we observed very low moose numbers in EINP-N compared to the other study areas due to mortality caused by liver flukes and winter ticks (Samuel, 2004, 2007; Shury et al., 2019).

Aerial ungulate survey density estimates

Aerial ungulate surveys are conducted in EINP annually attempting “total counts” (ASRD 2010) by surveying a series of east–west transects at 0.5-km intervals. Aerial surveys were conducted during winter each year, typically during November–January depending on weather conditions (snow cover, cloud cover, etc.). We extended the aerial survey to BPRA in winter 2019, corresponding to the BPRA camera-trap survey period. Ungulates are counted separately for each study area (EINP-N, EINP-S, BPRA). We calculated AUS density estimates for each focal species (bison, elk, moose), study area, and year by dividing the corresponding total count by the study area’s landcover in km2. Study area landcover was calculated using a 30 × 30 m resolution supervised Landsat 2015 raster. We excluded open water from the total area of each study area by subtracting cells/pixels with open water. Bison, elk, and moose AUS density estimates were compared with TIFC estimates from the same study year. We consider a study year to be the time period beginning on April 16th until the following April 15th (i.e., Study year 2017 includes April 16, 2017–April 15, 2018).

Camera trapping

We deployed camera traps (Reconyx Hyperfire: H500, P800, P900) on a 2 × 2 km2 systematic grid across all three sites with 31 cameras in EINP-N, 12 cameras in EINP-S, and 23 cameras in BPRA (Fig. 1). In places where a point on the grid fell within a lake or grassland, the camera was moved to the nearest tree. Each camera was attached to a tree 1 m off the ground, facing northward to prevent glare. Camera traps were placed facing areas where ungulate detection would be possible, such as game trails or open areas. Cameras on game trails were placed approximately 4 m away from the game trail with an unobstructed view of the game trail. In EINP-N and EINP-S, cameras that would have been placed on or near human-use trails were moved off trail by 100 m for privacy concerns. Cameras deployed in BPRA were not placed to avoid human-use trails. To maximize the chance of capturing fast-moving animals, cameras were set to take three rapid-fire photos with up to two frames per second following each motion trigger. Cameras were serviced twice each year at a minimum. We cleared vegetation within the field of view every time cameras were serviced.

For this study, we used camera-trap data collected in EINP-N and EINP-S from April 2017 to April 2020, and we used camera-trap data collected in BPRA from April 2019 to April 2020. The number of active cameras used for density estimation varied each year due to equipment removals and/or failures (Table 1).

Table 1.

Number of cameras (n), camera trapping days (CT days), and detection events for bison, elk, moose, mule deer, and white-tailed deer (WTD) in each of three study areas in central Alberta, Canada from 2017 to 2019

Study area* Study year n CT days Number of detection events
Bison Elk Moose Mule deer WTD
EINP-N 2017 15 6096 776 1196 65 70 595
2018 30 9245 551 2292 79 51 732
2019 29 10,824 1419 3064 81 68 983
EINP-S 2017 12 3747 618 879 634 76 1030
2018 10 2981 224 681 479 31 732
2019 10 3788 236 1160 547 23 1041
BPRA 2019 22 6421 - 989 535 180 783
Total 43,102 3824 10,261 2420 499 5896
Open in a new tab

*EINP-N, Elk Island National Park north; EINP-S, Elk Island National Park south; BPRA, Cooking Lake—Blackfoot Provincial Recreation Area

Photos were stored in the WildTrax image-tagging system (WildTrax, 2019), including metadata for each photo such as camera name, date, and temperature. Each photo was tagged with information for species name, age class, sex, and number of individuals following ABMI tagging protocols (WildTrax, 2019).

Time in front of the camera (TIFC) model

We calculated densities for each species at each camera station for each survey year. To estimate density D^ (animals/km2) using camera traps, we applied the TIFC model (Becker et al., 2022):

D^=∑N∙TFAF∙TO

where N is the number of animals observed, TF is the time spent in the field-of-view, AF is the area in the field of view, and TO is the camera operating time.

We defined independent events using an interval of 120 s between photo sequences. The TIFC approach uses the cumulative staying time for any member of a given species during the monitoring period (Becker et al., 2022; Warbington & Boyce, 2020). Staying time (N∙TF) was calculated per event by taking the average number of animals per photo in the event (N) multiplied by the event duration (TF). Because our data consisted of discrete time-stamped images, we modified the time spent in the field of view in two ways following the approach in Becker et al. (2022). First, we accounted for time spent in the field of view before the first photo and after the last photo by adding on a species-specific average time between photos to each event. Second, we adjusted the time spent in the field of view based on the time intervals between photos in each event. For gaps less than 20 s, we assumed the animal stayed in the field of view; for gaps greater than 120 s, we assume the animal left (and the next detection would be a new independent event); and for gaps between 20–120 s, we applied species-specific models for the probability of leaving (Becker et al., 2022). There was not a species-specific leaving-time model for bison, so for bison we used a model that incorporated data for multiple large ungulates (i.e., bison, elk, mule deer, white-tailed deer; Becker et al., 2022). Staying time for each species was then summed for all events per camera location, per survey period.

For camera operating time (TO) we used the number of camera trapping days sampled that season and study year. Area in the field of view (AF) was calculated using the following equation:

AF(m2)=(π×EDD2×angle)360

where EDD is the effective detection distance (m). Effective detection distances were estimated for each species, each habitat type, and each season (Becker et al., 2022). We assumed angle to be 42° for all cameras (Reconyx, 2017).

TIFC assumptions include (1) representative sampling of microhabitats, (2) animal movement is not altered by camera-trap presence, and (3) perfect detection within 5 m of the camera (Becker et al., 2022; Warbington & Boyce, 2020).

Estimating density

The TIFC equation was used to estimate density of a species at each camera location for a given time period. We defined two seasons: summer (April 16–October 15) and winter (October 16-April 15). The time frame for winter was modified for bison in the study areas and years when bison handling took place (Table 1, Appendix A). We used the TIFC method to estimate ungulate densities at each camera station for the summer and winter. We then estimated density for the study year by averaging the two seasons. This method estimates a mean density for the sampling period, similar to time-to-event models (Loonam et al., 2021), and does not rely on the strict assumption of population closure that is required for mark-recapture methods to avoid bias. Cameras with fewer than 20 days per season were excluded from analysis. We then applied species and habitat-specific correction factors to elk, moose, mule deer, and white-tailed deer estimates to account for the fact that their staying time is inflated by time spent investigating cameras, a violation of assumption 2 for which we attempted to adjust using correction factors (Becker et al., 2022). Lastly, we estimated densities per species, per study area, and per study year by averaging the density estimates at all cameras from the corresponding study area and time frame. We calculated 90% confidence intervals using a Monte Carlo simulation of both presence/absence and abundance given presence using the same methods applied in Becker et al. (2022).

Results

Aerial ungulate survey density estimates

We calculated land area for each of the three study areas: 128.5 km2 in EINP north, 55.5 km2 in EINP-south, and 91.2 km2 in BPRA. We used counts from the winter aerial surveys divided by land area (km2) of each study area to calculate AUS density estimates for bison, elk, and moose (Fig. 2, Table 2). The aerial survey in EINP-S in 2017 could not be completed in the usual timeframe (Dec/Jan) and the survey was split and completed during two different time periods. Due to these sampling issues, we omitted the aerial total counts from the EINP-S 2017 survey and did not calculate AUS density estimates for this study area and time period. There are no estimates of precision associated with aerial survey total counts.

Fig. 2.

Fig. 2

Open in a new tab

Bison, elk, moose, mule deer, and white-tailed deer density estimates from the Time-in-front-of-the-camera (TIFC) approach and Aerial Ungulate Survey density approach. Data were collected during three years across three study areas in Alberta, Canada: Elk Island National Park-North (EINP-N; 2017–2019), Elk Island National Park-South (EINP-S; 2017–2019), and Cooking Lake—Blackfoot Provincial Recreation Area (BPRA; 2019). Error bars represent 90% confidence intervals for the TIFC density estimates

Table 2.

Sample size (n), time-in-front-of-camera (TIFC) density estimates, TIFC 90% confidence intervals, and Aerial Ungulate Survey (AUS) density estimates for bison, elk, moose, mule deer, and white-tailed deer in three study areas in Alberta, Canada from 2017 to 2019

Species Study area* Year n Density per km2
TIFC estimate TIFC 90% CI AUS estimate
Bison (Bison bison) EINP-N 2017 15 5.41 1.73–12.01 4.17
2018 25 1.82 0.61–3.86 4.87
2019 29 6.19 2.66–11.56 4.13
EINP-S 2017 10 8.07 2.85–17.10 -
2018 10 2.54 1.44–4.06 6.48
2019 10 3.53 1.38–6.97 7.56
Elk (Cervus elaphus) EINP-N 2017 15 4.75 3.16–6.73 3.26
2018 30 5.03 3.88–6.39 2.56
2019 29 6.17 4.79–7.79 4.05
EINP-S 2017 12 7.87 3.90–13.58 -
2018 10 6.73 1.69–16.45 2.65
2019 10 8.71 3.38–17.39 3.22
BPRA 2019 22 3.55 1.68–6.35 1.27
Moose (Alces alces) EINP-N 2017 15 0.11 0.06–0.20 0.25
2018 30 0.06 0.03–0.09 0.16
2019 29 0.10 0.05–0.18 0.13
EINP-S 2017 12 5.46 3.32–8.33 -
2018 10 5.09 3.37–7.30 3.8
2019 10 3.61 2.07–5.75 2.75
BPRA 2019 22 1.36 0.56–2.61 1.5
Mule deer (Odocoileus hemionus) EINP-N 2017 15 0.25 0.06–0.60 -
2018 30 0.08 0.03–0.15 -
2019 29 0.10 0.04–0.18 -
EINP-S 2017 12 0.62 0.10–1.73 -
2018 10 0.15 0.04–0.34 -
2019 10 0.22 0.05–0.57 -
BPRA 2019 22 0.38 0.19–0.67 -
White-tailed deer (O. virginianus) EINP-N 2017 15 1.38 1.00–1.84 -
2018 30 1.37 0.97–1.87 -
2019 29 1.56 1.14–2.07 -
EINP-S 2017 12 5.63 4.54–6.86 -
2018 10 4.20 3.21–5.36 -
2019 10 5.30 3.88–7.01 -
BPRA 2019 22 1.39 0.90–2.00 -
Open in a new tab

*EINP-N, Elk Island National Park north; EINP-S, Elk Island National Park south; BPRA, Cooking Lake—Blackfoot Provincial Recreation Area

Camera trapping

In total, we included 43,102 camera-trapping days (Table 1). The minimum number of camera-trapping days for any study area and survey period was 2,981. We summarized the total number of capture events for each species for each study area and year (Table 1). The minimum number of capture events for any study area and year were 224 for bison, 681 for elk, 65 for moose, 23 for mule deer, and 595 for white-tailed deer.

TIFC density estimates

We calculated TIFC density estimates and 90% confidence intervals for bison, elk, moose, mule deer, and white-tailed deer for each study area and year (Fig. 2, Table 2).

Bison TIFC estimates ranged from 1.82 to 6.19 bison per km2 in EINP-N and from 2.54 to 8.07 bison per km2 in EINP-S (Table 2, Fig. 2). For study areas and years with both a camera density and AUS density estimates for comparison, bison TIFC estimates were lower than AUS density estimates in 3 out of 5 occasions (Fig. 3). Using these 5 occasions, bison TIFC density estimates were only 19% lower than bison AUS density estimates, with an average difference of 1.55 bison per km2.

Fig. 3.

Fig. 3

Open in a new tab

Time-in-front-of-the-camera (TIFC) density estimates compared to AUS density estimates for bison, elk, and moose in three study areas in Alberta, Canada: Elk Island National Park-North (EINP-N), Elk Island National Park-South (EINP-S), and Cooking Lake—Blackfoot Provincial Recreation Area (BPRA). Error bars represent 90% confidence intervals for the TIFC density estimates. The dashed line represents the 1:1 relationship between AUS density estimates and camera densities (animals per km2)

TIFC estimates for elk ranged from 3.55 to 8.71 elk per km2 across all study areas and years, with 90% confidence intervals overlapping for all estimates (Table 2, Fig. 2) indicating that elk density did not have a detectable difference across the three study areas. For study areas with multiple years (EINP-N and EINP-S), elk density estimates were consistent across years (Fig. 2). For study areas and years with corresponding camera and AUS density estimates, we found that elk TIFC density estimates were higher than AUS density estimates on all 6 occasions (Fig. 3). Using these 6 occasions, elk TIFC estimates were 116% higher than AUS density estimates, with an average difference of 2.99 elk per km2.

In EINP-N, moose TIFC estimates were the lowest of the three study areas with estimates ranging from 0.06 to 0.11 moose per km2 (Table 2). BPRA had an intermediate TIFC estimate at 1.36 moose per km2. EINP-S had the highest moose TIFC estimates ranging from 3.61 to 5.46 moose per km2. For study areas with multiple years (EINP-N and EINP-S), moose density estimates were consistent across years (Fig. 2). For study areas and years with corresponding camera and AUS density estimates, we found that moose TIFC density estimates aligned well with AUS density estimates in all 6 occasions (Fig. 3). Moose TIFC estimates were higher than moose AUS density estimates in EINP-S, but moose TIFC estimates were lower than moose AUS density estimates in EINP-N and BPRA. On average, moose TIFC estimates were higher than moose AUS density estimates by 0.29 moose per km2 due to the higher TIFC values in EINP-S, the study area with the highest moose density. We found that the proportional difference for moose was misleading because of the near-zero density estimates for moose in EINP-N. Moose TIFC density estimates in EINP-N were only lower than moose AUS density estimates by 0.09 moose per km2 on average, but due to the low densities, they had the highest proportional difference.

Mule deer TIFC density estimates were low across the three study areas, ranging from 0.08 to 0.62 mule deer per km2 (Table 2), and estimates were generally consistent across years. White-tailed deer TIFC density estimates ranged from 1.37 to 1.56 deer per km2 in EINP-N, with a similar density estimated in BPRA of 1.39 deer per km2. White-tailed deer TIFC estimates were higher in EINP-S, ranging from 4.20 to 5.63 deer per km2. In all study areas, white-tailed deer density estimates were higher than mule deer estimates, consistent with park staff observations.

Discussion

We applied the TIFC model to five species of ungulates in three enclosed study areas across multiple years. For the three species with AUS density estimates for comparison (i.e., bison, elk, and moose), we found that TIFC and AUS density estimates were most similar for moose, whereas bison and elk comparisons showed higher variation between methods (Fig. 2, Table 2).

Overall, Moose had the lowest discrepancy between TIFC and AUS density estimates, and across study areas and years, results were correlated (Fig. 3). TIFC estimates were higher than AUS density estimates by only 0.29 moose per km2 on average. Aerial surveys in our three study areas are considered to be fairly reliable for moose given the high sightability of moose within the habitat types present (Bisset & Rempel, 1991; Gasaway et al., 1986). Moose are highly visible in deciduous forests (83% seen), deciduous shrub (100% seen), and open habitats (100% seen) – the three habitat types that constitute over 95% of each of our study areas (Anderson & Lindzey, 1996). Both aerial and TIFC densities were realistic given our knowledge of moose populations in each study area. We observed consistently low densities (< 0.26 moose/km2) for both TIFC and aerial approaches in EINP-N. The moose population in EINP-N has been reduced by high mortality from winter ticks and liver flukes (Samuel, 2004, 2007; Shury et al., 2019), resulting in low moose densities compared to the other two study areas. In contrast, the moose population in EINP-S is high, resulting in management concerns due to overgrazing (Parks Canada, 2017). Using both TIFC and aerial surveys, we observed high moose densities (2.75–5.46 moose/km2) across years in EINP-S (Fig. 2). We predicted that BPRA would have intermediate values for moose density relative to EINP-N and EINP-S because the fenced perimeter limits dispersal, but Indigenous and licensed hunting are permitted in BPRA which reduces population numbers. As expected, we observed intermediate moose densities in BPRA, with TIFC densities and AUS density estimates that were similar (1.36 and 1.5 moose/km2). Although confidence intervals from TIFC estimates of moose density are large, these estimates were more precise when compared to elk and bison estimates across all three study areas regardless of the sample size (Fig. 2, Table 2). EINP-S had the largest confidence intervals, and precision for moose estimates in EINP-S would likely be improved by increasing the number of monitoring locations.

Bison TIFC densities were 19% lower than aerial survey densities, with an average difference of 1.55 bison per km2. We consider the AUS density estimates for bison to be reliable due to the high sightability of bison, with previous studies estimating bison sightability of 92–97% (Hess, 2002; Terletzky & Koons, 2016; Wolfe & Kimball, 1989). Bison sightability is high due to their large body size, gregarious behavior, and preferential use of open habitats like grasslands. Aerial counts for bison are probably close to the true population size in the park, and thus the TIFC bison estimates were biased low. This was likely due in part to inadequate sampling of open habitats like grasslands because we were unable to deploy cameras in the middle of fields using naturally occurring trees and constraints around using artificial posts in a national park. Post-hoc, we calculated the proportion of habitat types sampled compared to the proportion of habitats present in EINP-N and EINP-S and found that grasslands and other open habitats were underrepresented in both study areas (Appendix B), a violation of assumption 1 of the TIFC method (Becker et al., 2022). Systematic grids are a common arrangement used in camera trapping studies for density estimation; however, stratification by habitat type and/or increasing sampling effort should be considered for camera trapping studies, especially when monitoring multiple species with different patterns of habitat use in heterogenous landscapes. Future studies using TIFC should consider stratified random sampling to ensure the camera trap study design is representative in terms of availability of habitats and optimized relative to variance (Williams et al., 2002). Stratified random sampling has been identified as an option for TIFC to achieve representative sampling (Becker et al., 2022), and we believe stratification by habitat type would likely reduce bias for bison TIFC estimates. Another consideration for bison is that they are a highly gregarious species, resulting in higher variation in staying time calculations compared to moose. Larger camera sample sizes are therefore recommended to improve the precision of TIFC estimates for bison.

Elk estimates differed between the camera trap and aerial survey approaches with TIFC estimates that were 116% higher than aerial survey densities with a difference of 2.99 elk per km2 on average, indicating that elk TIFC estimates were likely biased high. While it is possible that elk AUS counts are lower than the true population values due to lower sightability for elk compared to bison and moose (Allen et al., 2008; Caughley, 1974; Samuel et al., 1987), we expect sightability in our study areas to be relatively high due to the low amount of conifer present. One potential source of bias for elk density estimates is the non-random selection of microhabitats for camera sampling locations. Placement of cameras on game trails or open areas can result in sampling areas with higher wildlife use than what is representative of the landscape, which would lead to density estimates that are biased high (Becker et al., 2022). Because ungulate density is very high in EINP and game trails are prevalent across the study areas, the degree to which camera placement may have resulted in TIFC estimates that are biased high is unknown. Further data collection using a combination of game trail and random deployments could potentially be used to calibrate the analysis (Becker et al., 2022). Another potential source of bias for elk is investigating behaviors, which can greatly increase staying times. While we did apply correction factors created by ABMI (Becker et al., 2022), it is possible that correction factors may need to be adjusted for different study areas. In our study areas, which consisted of three fenced areas, two of which do not allow hunting (i.e., ENIP-N and ENP-S), it is possible that time spent on investigating behaviors may vary across areas with different conditions. Additional assessment may be required to determine the exact cause of bias for elk TIFC estimates in our study areas and to determine the best way to reduce bias, either through modification of the sampling design or validation of the correction factors used for elk investigation behaviors. We saw the largest discrepancy between elk TIFC estimates and AUS density estimates in EINP-S, where elk TIFC estimates had high variance across years. The low degree of precision is likely due to the low number of cameras in EINP-S combined with elk having higher variation in detection probability and staying time due to variable group sizes. Similar to bison, accurate censuses of elk likely require more cameras relative to a solitary species like moose to improve the precision of estimates.

We did not have an aerial estimate of mule deer and white-tailed deer densities in our three study areas, so we could not compare deer TIFC estimates with AUS density estimates. However, using the TIFC model we were able to provide the only population estimates of mule deer and white-tailed deer park managers have in both EINP and BPRA. These estimates were consistent across years for study areas with multiple study years, and estimates were relatively precise (Fig. 2). The high precision of the deer estimates can be attributed to the relatively consistent estimates across cameras compared to the large range of density values for bison and elk at individual cameras.

Overall, the two sampling approaches were complementary. Aerial surveys in our study areas had excellent spatial coverage compared to the camera trap surveys. The number of camera traps available to researchers can limit the spatial coverage of camera surveys. Further, one limitation of the TIFC approach is that it can be difficult to determine if the camera trap array is collecting an adequate, representative sample, especially for multispecies monitoring. This issue is present in other methods for multi-species density estimation as well (Burton et al., 2015; Rowcliffe et al., 2008). To counter this issue, camera trap researchers should incorporate knowledge of species ecology into sampling design, rather than simply adopting a systematic grid or random sampling design.

While the aerial surveys excel at spatial coverage, they provided poor temporal coverage compared to cameras, with only a once-per-year snapshot in winter for each of our study areas. While this may not be important for annual population comparisons, the camera data has the potential to be applied for additional applications, including seasonal comparisons. Camera traps allowed us to identify a larger number of species compared to the AUS and also allowed us to collect detailed demographic information throughout the year. Both the timing of the winter aerial survey and sightability issues make it challenging to collect demographic data (i.e., sex ratios, age class distribution). Future studies could use demographic data to subset the data in a variety of ways prior to estimating density using TIFC. Future studies could also experiment with applying the TIFC approach for shorter time frames during the year, which may reduce the variance of TIFC estimates.

Another limitation for the aerial surveys applied in this study is that the “total count” approach used for surveying ungulates in EINP does not include estimates of precision. For future comparisons between camera surveys and aerial surveys, changing the aerial survey approach to distance sampling would be more appropriate to account for sightability and to produce estimates of precision. Understanding the precision of aerial survey density estimates would facilitate comparisons with TIFC estimates, which do include estimates of precision.

Other considerations for camera surveys versus AUS include that camera trapping is much more labor intensive. Both the field work for deploying and maintaining camera traps and for photo processing can take a tremendous amount of time. Advancements in species recognition or “auto-tagging” software have the potential to greatly reduce the time commitment required for processing photos (Glover-Kapfer et al., 2019).

A major strength of the TIFC approach is that it is more accessible than some of the other methods due to applicability for unmarked species and fewer data requirements. The TIFC method can be used to estimate densities for a broad range of species and demographic subgroups without needing individual recognition, movement rates, home range size, or average group size. In this study, we were able to estimate TIFC densities for focal species and non-focal species, such as white-tailed deer, mule deer, black bears, coyotes, and wild boar (Sus scrofa). Future research on the TIFC approach should investigate the effect of sample size and different sampling designs on the accuracy and precision of TIFC density estimates.

Conclusion

We have shown that the TIFC approach for density estimation produced complementary results to aerial surveys, with pros and cons of the two approaches varying based on focal species ecology. Cameras may be more appropriate for detecting species in closed habitats and species with lower sightability for aerial surveys (e.g., elk, white-tailed deer, mule deer), while aerial survey density estimates are potentially more appropriate for bison, a highly gregarious ungulate which uses primarily open habitats and has high sightability via aerial surveys. Careful consideration is required for several factors related to camera trap study designs for TIFC density estimation that may affect the accuracy and precision of estimates. These factors include how to meet the assumption of representative sampling through study design at both the landscape scale and microhabitat scale, which can affect the accuracy of estimates differently across species of interest. For study designs incorporating cameras placed on game trails, some calibration is likely required to understand relative species use between game trails versus random microhabitat features (Becker et al., 2022). Calibration may also be required to account for investigating behaviours, which may vary by species and study area. Lastly, several factors need to be considered to attempt to reduce measurement error for TIFC density estimates (Becker et al., 2022), including consideration of landscape heterogeneity, expected variation in habitat use for focal species, and whether focal species are gregarious. Highly heterogenous landscapes, species that are more likely to show strong selection for specific habitats relative to their availability on the landscape, and highly gregarious species are all likely to result in high variation in camera staying times across sampling locations, which can result in imprecise TIFC estimates. Larger camera sample sizes, longer deployment times, and estimating density per habitat strata are all options that may improve the precision of TIFC estimates.

Wildlife managers often are limited by resources and funding, creating a need for cost-effective methods to monitor multiple species. Wildlife managers should apply careful consideration for survey type and analysis method for density estimation, while balancing costs and competing monitoring needs. Camera traps and the TIFC approach may be a viable option if wildlife managers are able to meet TIFC model assumptions and achieve adequate sample sizes, with additional benefits of surveying a broad range of species, capturing seasonal variation, and collecting demographic data, all of which are limited using aerial surveys.

Acknowledgements

We thank L. Roy and the Friends of Elk Island Society for help with project design, assisting with grant acquisition, and supporting the field work. We thank Parks Canada for their logistical support and field work support. Thanks to J. Singh, M. Brownlee, K. Lanzenstiel, J. Pattinson, T. Strydhorst, S. Widmeyer, and R. Dawson for their help with data collection and data processing. Additionally, we thank M. Eklund, A. Zapata, J. Benade, C. Brooke, E. Hollik, H. Holzer, R. Martins, C. Nobleza, A. Suleman, and C. Zhao for their help with data processing. Research was permitted by Parks Canada (EI-2016-23336, EI-2019-31317, EI-2020-35818).

Author contributions

JF, MSB, and DV contributed to study conception and design; JF and DV collected the data; JF and MB contributed to data analysis; JF, MSB, DV, and MB contributed to manuscript preparation. All authors contributed critically to the drafts and gave final approval for publication.

Funding

This project was financially supported by the University of Alberta, an Alberta Conservation Association grant to L. Roy and the Friends of Elk Island Society (030–00-90–261), the Safari Club International Northern Alberta Chapter—Hunting Heritage Fund grants to MSB (RES0044334, RES0049275), the Alberta Fish and Game Association – Minister’s Special Licence grants to MSB (RES0044700, RES0049598), and the Beaver Hills Initiative Grant to DV. Funds for open access publication were provided by the Natural Sciences and Engineering Research Council of Canada grant to MSB (RES0061179).

Data availability

Camera-trap data collected and analyzed for this study are publicly available through WildTrax at https://portal.wildtrax.ca/home/projects/data-download.html.

Declarations

Conflict of  interests

The authors have no competing interests to declare that are relevant to the content of this article.

Footnotes

Publisher's note

Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.

References

  1. Alberta Sustainable Resource Development (ASRD).(2010). Aerial ungulate survey protocol manual. Alberta Sustainable Resource Development Fish and Wildlife Division, Alberta. https://open.alberta.ca/publications/aerial-ungulate-survey-protocol-manual/resource/1604125e-408e-40af-9bda-091ff6a25dde. Accessed 08 Jul 2021.
  2. Allen, J. R., Mcinenly, L. E., Merrill, E. H., & Boyce, M. S. (2008). Using resource selection functions to improve estimation of elk population numbers. The Journal of Wildlife Management,72(8), 1798–1804. 10.2193/2007-311 [Google Scholar]
  3. Anderson Jr, C. R., & Lindzey, F. G. (1996). Moose sightability model developed from helicopter surveys. Wildlife Society Bulletin, 24(2), 247–259. https://www.jstor.org/stable/3783114
  4. Becker, M., Huggard, D. J., Dickie, M., Warbington, C., Schieck, J., Herdman, E., Serrouya, R., & Boutin, S. (2022). Applying and testing a novel method to estimate animal density from motion-triggered cameras. Ecosphere,13(4), Article e4005. 10.1002/ecs2.4005 [Google Scholar]
  5. Bender, L. C. (2006). Uses of herd composition and age ratios in ungulate management. Wildlife Society Bulletin,34(4), 1225–1230. 10.2193/0091-7648(2006)34[1225:UOHCAA]2.0.CO;2 [Google Scholar]
  6. Bisset, A. R., & Rempel, R. S. (1991). Linear analysis of factors affecting the accuracy of moose aerial inventories. Alces,27, 127–139. [Google Scholar]
  7. Borchers, D. L., & Efford, M. G. (2008). Spatially explicit maximum likelihood methods for capture–recapture studies. Biometrics,64(2), 377–385. 10.1111/j.1541-0420.2007.00927.x [DOI] [PubMed] [Google Scholar]
  8. Boyce, M. S., Baxter, P. W., & Possingham, H. P. (2012). Managing moose harvests by the seat of your pants. Theoretical Population Biology,82(4), 340–347. 10.1016/j.tpb.2012.03.002 [DOI] [PubMed] [Google Scholar]
  9. Burton, A. C., Neilson, E., Moreira, D., Ladle, A., Steenweg, R., Fisher, J. T., Bayne, E., & Boutin, S. (2015). Wildlife camera trapping: A review and recommendations for linking surveys to ecological processes. Journal of Applied Ecology,52(3), 675–685. 10.1111/1365-2664.12432 [Google Scholar]
  10. Caughley, G. (1974). Bias in aerial survey. The Journal of Wildlife Management,38(4), 921–933. 10.2307/3800067 [Google Scholar]
  11. Caughley, G., Sinclair, R., & Scott-Kemmis, D. (1976). Experiments in aerial survey. The Journal of Wildlife Management,40(2), 290–300. 10.2307/3800428 [Google Scholar]
  12. Chandler, R. B., & Royle, J. A. (2013). Spatially explicit models for inference about density in unmarked or partially marked populations. The Annals of Applied Statistics. 10.1214/12-AOAS610 [Google Scholar]
  13. Efford, M. (2004). Density estimation in live-trapping studies. Oikos,106(3), 598–610. 10.1111/j.0030-1299.2004.13043.x [Google Scholar]
  14. Foca, J. M., & Boyce, M. S. (2022). Bison and elk spatiotemporal interactions in Elk Island National Park. Frontiers in Conservation Science,3, Article 937203. 10.3389/fcosc.2022.937203 [Google Scholar]
  15. Foster, R. J., & Harmsen, B. J. (2012). A critique of density estimation from camera-trap data. The Journal of Wildlife Management,76(2), 224–236. 10.1002/jwmg.275 [Google Scholar]
  16. Gasaway, W. C., DuBois, S. D., Reed, D. J., & Harbo, S. J. (1986). Estimating moose population parameters from aerial surveys. University of Alaska. [Google Scholar]
  17. Glover-Kapfer, P., Soto-Navarro, C. A., & Wearn, O. R. (2019). Camera-trapping version 3.0: Current constraints and future priorities for development. Remote Sensing in Ecology and Conservation,5(3), 209–223. 10.1002/rse2.106 [Google Scholar]
  18. Graham, A., & Bell, R. (1989). Investigating observer bias in aerial survey by simultaneous double-counts. The Journal of Wildlife Management,53(4), 1009–1016. 10.2307/3809603 [Google Scholar]
  19. Habib, T. J., Moore, D. A., & Merrill, E. H. (2012). Detection and stratification approaches for aerial surveys of deer in prairie–parklands. Wildlife Research,39(7), 593–602. 10.1071/WR11175 [Google Scholar]
  20. Hess, S. C. (2002). Aerial survey methodology for bison population estimation in Yellowstone National Park (Thesis). Montana State University, Bozeman, MT.
  21. Karanth, K. U., & Nichols, J. D. (1998). Estimation of tiger densities in India using photographic captures and recaptures. Ecology,79(8), 2852–2862. 10.1890/0012-9658(1998)079[2852:EOTDII]2.0.CO;2 [Google Scholar]
  22. Krausman, P. R., & Bleich, V. C. (2013). Conservation and management of ungulates in North America. International Journal of Environmental Studies,70(3), 372–382. 10.1080/00207233.2013.804748 [Google Scholar]
  23. Loonam, K. E., Lukacs, P. M., Ausband, D. E., Mitchell, M. S., & Robinson, H. S. (2021). Assessing the robustness of time-to-event models for estimating unmarked wildlife abundance using remote cameras. Ecological Applications,31(6), Article e02388. [DOI] [PubMed] [Google Scholar]
  24. Lynch, G. M., & Shumaker, G. E. (1995). GPS and GIS assisted moose surveys. Alces 31:145–151.
  25. Mahoney, S. P., & Cobb, D. (2010). Future challenges to the model: Why collapse is possible and alteration inevitable. The Wildlife Professional,4, 83–85. [Google Scholar]
  26. McIntosh, T. E., Rosatte, R. C., Hamr, J., & Murray, D. L. (2009). Development of a sightability model for low-density elk populations in Ontario, Canada. The Journal of Wildlife Management,73(4), 580–585. 10.2193/2007-550 [Google Scholar]
  27. Moeller, A. K., Lukacs, P. M., & Horne, J. S. (2018). Three novel methods to estimate abundance of unmarked animals using remote cameras. Ecosphere,9(8), Article e02331. 10.1002/ecs2.2331 [Google Scholar]
  28. Nakashima, Y., Fukasawa, K., & Samejima, H. (2018). Estimating animal density without individual recognition using information derivable exclusively from camera traps. Journal of Applied Ecology,55(2), 735–744. 10.1111/1365-2664.13059 [Google Scholar]
  29. Parks Canada. (2010). Elk Island National Park: State of the park report (R61-29/2010E-PDF). Elk Island National Park, Fort Saskatchewan.
  30. Parks Canada. (2017). What we heard: Ungulate management in Elk Island National Park - October 2016 to June 2017. Elk Island National Park, Fort Saskatchewan. https://parks.canada.ca/pn-np/ab/elkisland/nature/conservation/faune-wildlife/ongules-ungulate2
  31. Reconyx. (2017). Reconyx hyperfire high performance cameras instruction manual. Reconyx. https://images.reconyx.com/file/HyperFireManual.pdf. Accessed 1 Jul 2021.
  32. Rowcliffe, J. M., Field, J., Turvey, S. T., & Carbone, C. (2008). Estimating animal density using camera traps without the need for individual recognition. Journal of Applied Ecology. 10.1111/j.1365-2664.2008.01473.x [Google Scholar]
  33. Royle, J. A., Chandler, R. B., Sollmann, R., & Gardner, B. (2013). Spatial capture-recapture. Waltham, MA: Academic press. [Google Scholar]
  34. Royle, J. A., Chandler, R. B., Sollmann, R., & Gardner, B. (2014). Spatial capture-recapture. Waltham: Academic Press. [Google Scholar]
  35. Samuel, B. (2004). White as a ghost: Winter ticks & moose. Edmonton, AB: Natural History Federation of Alberta Naturalists. [Google Scholar]
  36. Samuel, M. D., Garton, E. O., Schlegel, M. W., & Carson, R. G. (1987). Visibility bias during aerial surveys of elk in northcentral Idaho. The Journal of Wildlife Management,51(3), 622–630. 10.2307/3801280 [Google Scholar]
  37. Samuel, W. M. (2007). Factors affecting epizootics of winter ticks and mortality of moose. Alces,43, 39–48. [Google Scholar]
  38. Sasse, D. B. (2003). Job-related mortality of wildlife workers in the United States, 1937–2000. Wildlife Society Bulletin,31(4), 1015–1020. [Google Scholar]
  39. Shury, T. K., Pybus, M. J., Nation, N., Cool, N. L., & Rettie, W. J. (2019). Fascioloides magna in moose (Alces alces) from Elk Island National Park, Alberta. Veterinary Pathology,56(3), 476–485. 10.1177/0300985818823776 [DOI] [PubMed] [Google Scholar]
  40. Steinhorst, R. K., & Samuel, M. D. (1989). Sightability adjustment methods for aerial surveys of wildlife populations. Biometrics,45(2), 415–425. 10.2307/2531486 [Google Scholar]
  41. Terletzky, P. A., & Koons, D. N. (2016). Estimating ungulate abundance while accounting for multiple sources of observation error. Wildlife Society Bulletin,40(3), 525–536. 10.1002/wsb.672 [Google Scholar]
  42. Warbington, C. H., & Boyce, M. S. (2020). Population density of sitatunga in riverine wetland habitats. Global Ecology and Conservation,24, Article e01212. 10.1016/j.gecco.2020.e01212 [Google Scholar]
  43. WildTrax. (2019). User Guide: Remote camera trap image tagging. Version 2019–10–10. Alberta Biodiversity Monitoring Institute.https://www.wildtrax.ca . Accessed 31 Dec 2019.
  44. Williams, B. K., Nichols, J. D., & Conroy, M. J. (2002). Analysis and management of animal populations. San Diego, CA: Academic press. [Google Scholar]
  45. Wolfe, M. L., & Kimball, J. F. (1989). Comparison of bison population estimates with a total count. The Journal of Wildlife Management,53(3), 593–596. 10.2307/3809181 [Google Scholar]

Associated Data

This section collects any data citations, data availability statements, or supplementary materials included in this article.

Data Availability Statement

Camera-trap data collected and analyzed for this study are publicly available through WildTrax at https://portal.wildtrax.ca/home/projects/data-download.html.

Articles from Environmental Monitoring and Assessment are provided here courtesy of Springer

RESOURCES