Abstract
Camera trapping has evolved from a specialist research technique to the standard platform for non-invasive wildlife
monitoring across habitats globally, with applications spanning species detection, occupancy estimation, abundance
indexing, behaviour analysis, and machine learning-based automated species identification. This study evaluates camera
trapping performance, design optimisation, and data analysis approaches for four wildlife monitoring objectives across
boreal (Finland), alpine (Austria), and continental (Denmark) habitats using 312 camera stations deployed for 84,621
trap-nights (2022-2024). Performance metrics assessed across 48 target species include detection probability,
occupancy estimation precision, and population density accuracy (via random encounter model, REM). Station spacing
optimisation experiments (24 configurations tested) demonstrate that species-specific activity range determines optimal
inter-station distance: 500-800 m spacing maximised small mammal detection rates while 1,500-2,500 m spacing was
optimal for large carnivores. Machine learning classification using a fine-tuned EfficientNetV2 model achieved 94.8 +-
2.4% species identification accuracy across 38 species for images with adequate contrast -- reducing manual image
processing time by 78.4%. Random encounter model density estimates for roe deer (Capreolus capreolus) correlated
strongly with independent GPS-telemetry density estimates (r = 0.88, p < 0.001). Deployment of camera trap networks
for Natura 2000 monitoring targets demonstrates that standardised protocols achieve sufficient statistical power (80%
power) to detect 20% occupancy change with 24-36 cameras deployed for 90 trap-nights. These results provide
evidence-based guidance for camera trap network design under EU Habitats Directive and national biodiversity
monitoring programme requirements.