Inferring predator-prey interactions from camera traps: A Bayesian co-abundance modeling approach

Area of Science:

• Ecology
• Wildlife population dynamics
• Conservation biology

Background:

• Studying predator-prey interactions is crucial but challenging due to difficulties in direct observation.
• Camera trapping provides extensive wildlife data, yet inferring species interactions remains complex.
• Existing co-abundance models struggle with species exhibiting vastly different densities and detection rates, common in predator-prey systems.

Purpose of the Study:

• To develop a novel Bayesian hierarchical N-mixture co-abundance model capable of inferring predator-prey interactions.
• To address limitations of current models in handling zero-inflation and overdispersion in count data typical of predator-prey dynamics.
• To provide a robust framework for analyzing camera trap data to understand species interactions.

Main Methods:

• Developed a Bayesian hierarchical N-mixture co-abundance model incorporating an informed zero-inflated Poisson distribution for abundance.
• Included random effects in detection probability to account for overdispersion across sampling units and occasions.
• Validated the model using 20 camera trapping datasets from Southeast Asian tropical forests.

Main Results:

• The proposed model demonstrated superior performance, improved goodness-of-fit, and overcame convergence issues compared to alternative methods.
• Estimated predator-prey relationships revealed tigers negatively impacted muntjac abundance, supporting top-down regulation.
• Clouded leopards showed a positive association with muntjac and sambar deer, potentially due to shared responses to unmodeled factors.

Conclusions:

• The developed Bayesian co-abundance model effectively quantifies predator-prey relationships, even with significant differences in species abundance and detection.
• This approach offers broad applicability across diverse ecosystems and sampling strategies for ecological research.
• The methodology can aid in predicting cascading effects of predator population changes on food webs.