Moose (Alces alces) are a photogenic species found across boreal and semi-boreal forests of the Northern Hemisphere. Previous studies have used demographic data from remote cameras to estimate demographic parameters and population dynamics. A primary assumption is that these age and sex classes are accurately classified. However, numerous factors can influence the ability of observers to identify age and sex classes of moose captured on cameras. We used data from 84 cameras from a 3-year period (19 Nov 2021–30 Apr 2024) in northern Maine, USA, to evaluate how temporal, environmental, site-level, and endogenous factors influence the ability of observers to classify age and sex classes of moose. Using Bayesian categorical regression models, we found that temporal variability, position and proximity of moose from cameras, and the behavior of moose influenced our ability to identify age and sex classes. This information can be used to decide which periods to use data for population modeling and how to design studies to reduce the amount of uncertainty associated with different age and sex classes. We anticipate that our approach could also be used for other species whose age and sex classes can be differentiated using remote cameras.

Study Area

We used data from remote cameras located in Wildlife Management District (WMD) 4 in northwestern Maine along the border of Quebec, Canada. The landcover in WMD 4 was comprised of Laurentian–Acadian Northern Hardwood Forest and Acadian Low Elevation Spruce-Fir-Hardwood Forest. Commercial timber harvest is common throughout this region in both landcover types. Moose densities are highest in WMD 4 (4 moose/km²) compared to other WMDs in the state and WMD 4 is part of an experimental unit to decrease moose densities via harvest. The climate of this region is variable with cold and snowy winters and warm and humid summers.

Camera surveys

We used a stratified random design to capture the local and landscape gradients across the study area, setting two remote camera arrays that each had a total of 84 sites. We positioned one camera trap per site, facing north on a tree, 1–2 m above the ground, and pointed at a slight downward angle towards a snow stake positioned 5–7 m from the camera. Skunk essence and turkey feathers were used as attractants and placed directly on the stakes. We used a combination of infrared no-glow cameras, including Bushnell Core DS-4K No Glow (Bushnell Outdoor Products, Overland Park, KS), Reconyx Hyperfire 2 Pro (Reconyx Inc., Holmen, WI, USA), and Browning Recon Force Elite HP4 (Prometheus Group LLC, Birmingham, AL, USA) that were all programmed to take 1 picture when triggered with a trigger reset of 1 second. Cameras were visited once per season (4 times/year) to download data, refresh lures, and check that the cameras and snow stakes were correctly functioning.

Demographic Classification Process

We used the WildTrax platform (https://wildtrax.ca) to tag and organize the data. First, we recorded the number of individuals and classified the age and sex of all moose detected in photographs; all observers were trained using an established protocol to differentiate between the age and sex class of moose. We classified juveniles (1–12 months old) following a birth pulse calendar that began on 1 May of each year (the beginning of May is the earliest known parturition date in Maine) and ended on 30 April of the following year. We differentiated juveniles from adults (>12 months old) based on their morphological features (smaller size, lighter coloration, square body shape, and shorter snout). Similarly, we differentiated between males and females by assessing morphology of each animal (males have antlers and pedicles, do not have a white patch on their rump, and often have a larger dewlap, darker nose bridge, and larger size). If the age and sex classes were indistinguishable, we classified them as unknown and flagged them for review. For the next stage, we performed a verification of all pictures tagged as moose. First, we reviewed each picture to ensure that there were no false positives (e.g., deer tagged as a moose) and false negatives (i.e., no animals present in pictures) and then we verified if the age and sex classes were classified correctly. For pictures where the age or sex class could not be classified, we tagged these as unknown. The final decisions on ambiguous age and sex classes were made by two observers with advanced experience identifying the age and sex classes of moose. In total, there were 9 different age and sex class combinations, including definitive and unknown classifications: a) “Adult Female”, b) “Adult Male”, c) “Adult Unknown” (unknown sex class), d) “Juvenile Female”, e) “Juvenile Male”, f) “Juvenile Unknown” (unknown sex class), g) “Unknown Female” (unknown age class), h) “Unknown Male” (unknown age class), and i) “Unknown Unknown” (unknown age and sex class) (Fig. 2). Lastly, we organized pictures of moose that were detected within a 10-minute period into photographic clusters and used these as a response variable in our demographic models; anything greater than this threshold was considered a separate detection event.

Predictor Variables

The predictor variables were from the four groups that we hypothesized influenced the ability of observers to classify age and sex classes of moose (temporal, environmental, site-level, and endogenous). We obtained temporal variables from the date and time stamps of the first picture within a photographic cluster. We derived environmental variables using the ‘daymetr’ R package, extracting daily weather data (precipitation, water vapor pressure, snow water equivalent) for each camera location. Site-level variables such as proximity (i.e., the distance animals were detected from camera) and centrality (i.e., the position that moose were located within picture view frame) were calculated using the MegaDetector (MD) bounding boxes that are integrated within WildTrax and averaged across photographic clusters. Lastly, we obtained the endogenous variables by calculating the number of moose within a picture cluster (a categorical variable where 1 moose = “One” and >1 moose = “Several”), the total number of photographs within a picture cluster (an index of behavior), and the speed at which animals moved across the camera view frame. Speed was calculated by obtaining the standard deviation of MD bounding box locations within a picture cluster and dividing it by the time each moose spent within the view frame. We centered all continuous covariates using the scale function in ‘base’ R.