Data from: Using camera traps to estimate habitat preferences and occupancy patterns of vertebrates in boreal wetlands

Camera trap survey We conducted camera trap surveys at 50 wetlands over two summers in 2018 and 2019, using three cameras per pond in a triangular configuration. We used scent lures as attractants, and cameras were rotated and baited accordingly. Data collection occurred during two sessions of seven consecutive days per year. Each camera array consisted of three infrared Bushnell Trophy Cam HD motion-sensing digital cameras set to be active 24 hours/day. Cameras were placed at the edge of the pond and secured to a tree or a wooden stake at an average height of 30–60 cm at about 2–5 m from the water. Cameras were triggered by animal movements and programmed to take three photographs per trigger event, and a following video of 10-s, with a 1-min interval delay between detections to avoid that a single animal would be the subject of a long event of recording. We placed scent lure on a stick at 2 -5 m in front of the camera trap to increase the probability of detecting animals approaching the lure. At the end of each session, cameras were checked, photos were analyzed to identify bird and mammal species, and records from the same species at the same pond on the same day were combined into one detection event. Covariates such as pond type, forest cover, year, latitude, and distance to the nearest road were considered in analyzing bird and mammal occupancy and detection probability. Variables such as cumulative rainfall, days since snowmelt, and sampling effort were also factored into the detection analysis. Data processing and analysis To enhance species detection, we combined observations from three camera traps at a specific site on a given day, resulting in a data matrix of 100 rows and 14 columns per species. Using a multispecies occupancy framework, we analyzed bird and mammal communities, considering various factors such as pond type, forest cover, year, latitude, and distance to the nearest road. Model parameters were estimated using a Bayesian approach with Markov chain Monte Carlo in JAGS 4.3.0 within R 4.1.2. Convergence of the chains was assessed through trace plots, posterior density plots, and the Brooks-Gelman-Rubin statistic. Model fit was evaluated using posterior predictive checks and the area under the receiver operating characteristic curve.