Pairing field methods to improve inference in wildlife surveys while accommodating detection covariance

Area of Science:

• Wildlife ecology
• Ecological modeling
• Conservation biology

Background:

• Multiple field sampling methods are common in wildlife surveys for comparing efficacy, integrating data, and evaluating biases.
• Existing models assume independent detections between methods, which is often violated in practice when methods are used simultaneously or in close proximity.
• This assumption can lead to biased parameter estimates and underestimated uncertainty in wildlife occurrence and abundance.

Purpose of the Study:

• To develop new occupancy and spatial capture-recapture models that allow for covariance between detections from different methods.
• To compare the performance of these new models against traditional independence-assuming models using simulation.
• To empirically apply the models to American marten (Martes americana) survey data from remote cameras, hair catches, and snow tracking.

Main Methods:

• Development of occupancy and spatial capture-recapture models incorporating detection covariance.
• Simulation studies to assess estimator performance under independence and dependence scenarios.
• Empirical analysis of American marten data collected via paired remote cameras, hair catches, and snow tracking.

Main Results:

• Simulation results show that models assuming independence produce biased estimates and underestimate uncertainty when detections are correlated.
• The reformulated models are robust to both methodological independence and covariance.
• Empirical findings indicate remote cameras and snow tracking have similar detection probabilities for martens, but snow tracking yields false positives; cameras outperform hair catches for individual detection.

Conclusions:

• The developed models enhance the robustness of analyses that integrate multiple data sources.
• Accounting for detection covariance improves ecological inference and understanding of survey method reliability.
• The study highlights potential biases in distribution estimates from snow tracking and detection competition in hair catches.