Materials and Methods
The study regions covered altogether 19 known bean goose peatlands across Finland in 2020 and 2021 (Fig. 1). The peatlands were located in the provinces of North Karelia (N = 5), Northern Ostrobothnia (N = 8) and Lapland (N = 6). Fifty-six and fifty-three camera traps were deployed in 2020 and 2021, respectively, for the duration of the breeding season from the beginning of May to August, covering the season when the geese were nesting, caring for the offspring and moulting. Study locations were divided into 6.25 ha grids and one or two grids were used for camera trap monitoring, depending on size of the peatland (see Nykänen et al., 2021). In each study location, 2–4 game cameras were placed by experts specialized in bean goose ecology on peatlands with ponds/water bodies. Motion sensitive cameras were set to capture two still images with a 10 s delay between triggers, and time lapse cameras were programmed to take two still images every 15 minutes. Cameras were attached on top of each other (time lapse above the motion sensor camera) to a tree or wooden pole at a height of 1 m above the ground. Cameras were typically visited once or twice during the study period to replace the memory cards and batteries, if needed. After each field season, adult taiga bean geese and goslings were detected and counted manually from the images (Fig. 2).
No permits concerning animal welfare/ethics for bean goose camera trapping were required as birds were not intentionally approached, and remote camera traps are considered to be a non-invasive study method. Camera traps were set with landowners’ permission on private land and on state-owned areas with Metsähallitus permit (permit number MH1145/2018).
Generalized linear mixed modelling (GLMM) approach was used to investigate whether there was a difference in total taiga bean goose numbers (adults and goslings added together) captured by cameras triggered by motion sensor vs cameras set to time lapse. The GLMM was run in R (R Core Team 2023) using the package glmmTMB (Brooks et al. 2017) with the factors ‘trigger type’ (motion sensor and time lapse) and ‘year’ (2020 and 2021) as fixed factors and with the factor ‘site’ (19 individual peatland ponds) included as a random intercept in the model. The varying effort, resulting from the different number of cameras deployed on the study sites and the amount of time that the cameras were recording over the study period, was accounted for by including an offset-term in the model that was calculated by totalling the number of time periods of recording per trigger type for each day. Due to the data including a large number of zero values (some days and time-periods had zero goose counts), we ran different candidate models with a Tweedie, Poisson and two types of negative Binomial distributions (NB1 and NB2 parameterizations; Hardin and Hilbe 2018, Bolker 2022) with and without accounting for the zero-inflation.
In addition, we ran a set of logistic models (see Table 1) to investigate whether the camera trigger type affects the daily capture probability (presence or absence in photos) of taiga bean geese. Here, we define capture probability as “the probability that an animal is captured in a photo given it is present in the camera’s viewshed” as per Moeller et al. (2023). Since it was not possible to include the offset-term in this type of model, we added the camera effort as a continuous covariate in the models. Other covariates were included as fixed (‘trigger type’ and ‘year’) or random (‘site’) factors the same way as in the count models.
We then compared model fits using Akaike’s Information Criterion (AIC) values to determine the best fitting models. The goodness of fit of the models were assessed by creating scaled quantile residual plots via simulation using the R-package DHARMa (Hartig 2022).