• artificial intelligence
• behavioral trade-off
• camera trap
• facilitation
• mesocarnivores
• occupancy
• recolonization
• segregation
• spatiotemporal interaction
• structural equartion
• top-down
• trigonometric model
• wolf

Predators regulate ecosystems not only through predation, but also via non-consumptive (or behaviorally-mediated) effects that force coexisting species to balance risk and resource use. Non-invasive monitoring methodologies, such as camera traps, can expose these pathways, yet their data are noisy, massive and often analyzed with over-simplified statistical methods. Using the recent recovery of the grey wolf (Canis lupus) in the Apuan Alps (Italy) as a case study, we combined a critical assessment of existing analytical practices with novel methodological advances to reveal how top-down forces reverberate through a Mediterranean carnivore guild. A literature review revealed that most camera trap interaction studies still rely on pairwise occupancy or coarse temporal overlap coefficients, seldom addressing imperfect detection and rarely integrating spatial, temporal and trophic axes of the ecological niches. Addressing these gaps, we first reconstructed wolf recolonization of the study area from > 10 years of opportunistic records (2007-2020). We fitted a Bayesian occupancy model correcting for preferential sampling to camera trap data and estimated the number of territorial groups from playback howling surveys, observing the expected lag and expansion phases typical of wolf recolonizations, and confirming the establishment of a local population. To process one year of systematic imagery (2021-2022), we developed an user-friendly, R-based workflow implementing artificial intelligence, cutting species-level labelling time by two-thirds, and clearing a major bottleneck for the analysis of “big data”. We then quantified fine-scale interactions among wolves, red foxes (Vulpes vulpes), badgers (Meles meles) and martens (Martes spp.). We integrated weekly multi-season occupancy estimates – and their associated measurement error – into a structural equation model (SEM) that simultaneously assessed habitat drivers and the hypothesized directional biotic paths. Subsequently, we implemented hierarchical trigonometric generalized linear mixed models to estimate diel locomotory activity patterns of subordinate species and assess whether same-day detection of a dominant competitor determined shifts in the amplitude and timing of the activity peaks. Collectively, the results of spatial and temporal analyses revealed a facilitation-suppression mosaic driven by wolf presence, in which subordinate species implemented behavioral trade-offs between predation risk and exploitation of resources provided by the top predator, such as carrion. The wolf-red fox relationship was dominated by facilitation, with spatial association and subtle temporal displacement likely providing enough distance to limit the risk of dangerous encounters while preserving access to carrion subsides. Wolves suppressed badgers only spatially, and had an overall neutral effect on martens. Positive red fox and badger co-occurrence and heightened fox locomotory activity upon badger detection suggest resource-mediated commensalism or fox facilitation (possibly around earthworm diggings), and a certain degree of tolerance by the large mustelid. Finally, martens were not influenced by foxes in space, but responded to fox detection by increasing their activity at relatively safe times, likely tracking foxes to access shared resources. These spatiotemporal nuances – that may have not emerged under conventional pairwise or single-axis analyses – indicate that top predators presence creates a patchy landscape of risk and reward that subordinates exploit through behavioral trade-offs, in compliance with the landscape of fear (wolf displacement of badger) and fatal attraction (wolf-fox relationship, and partially fox-martens) hypotheses.