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

Abstract

It is common to use multiple field sampling methods when implementing wildlife surveys to compare method efficacy or cost efficiency, integrate distinct pieces of information provided by separate methods, or evaluate method-specific biases and misclassification error. Existing models that combine information from multiple field methods or sampling devices permit rigorous comparison of method-specific detection parameters, enable estimation of additional parameters such as false-positive detection probability, and improve occurrence or abundance estimates, but with the assumption that the separate sampling methods produce detections independently of one another. This assumption is tenuous if methods are paired or deployed in close proximity simultaneously, a common practice that reduces the additional effort required to implement multiple methods and reduces the risk that differences between method-specific detection parameters are confounded by other environmental factors. We develop occupancy and spatial capture-recapture models that permit covariance between the detections produced by different methods, use simulation to compare estimator performance of the new models to models assuming independence, and provide an empirical application based on American marten (Martes americana) surveys using paired remote cameras, hair catches, and snow tracking. Simulation results indicate existing models that assume that methods independently detect organisms produce biased parameter estimates and substantially understate estimate uncertainty when this assumption is violated, while our reformulated models are robust to either methodological independence or covariance. Empirical results suggested that remote cameras and snow tracking had comparable probability of detecting present martens, but that snow tracking also produced false-positive marten detections that could potentially substantially bias distribution estimates if not corrected for. Remote cameras detected marten individuals more readily than passive hair catches. Inability to photographically distinguish individual sex did not appear to induce negative bias in camera density estimates; instead, hair catches appeared to produce detection competition between individuals that may have been a source of negative bias. Our model reformulations broaden the range of circumstances in which analyses incorporating multiple sources of information can be robustly used, and our empirical results demonstrate that using multiple field-methods can enhance inferences regarding ecological parameters of interest and improve understanding of how reliably survey methods sample these parameters.

Incorporating citizen science data in spatially explicit integrated population models

“…However, non‐independent detection processes can occur, especially when co‐location or induced (trap happy/shy) behaviors make detection by multiple sampling devices more or less dependent on each other (Clare et al. ). Formulations for dependent detection processes similar to multi‐method occupancy models (Nichols et al.…”

The recent past and promising future for data integration methods to estimate species’ distributions

“…() also exploit this relationship, although indirectly through correlation structure rather than a joint parameterization structure. Clare, McKinney, Depue, and Loftin () address another key issue for data integration, which is when observations from two datasets are not independent.…”

Integrating multiple genetic detection methods to estimate population density of social and territorial carnivores

“…For instance, estimates of r between hair trapping and radio-collar telemetry data from brown bears (Ursus arctos) were found to be inconsistent (Tenan et al 2017), whereas r estimates between camera trapping and telemetry data from fishers (Pekania pennanti) were congruous (Popescu et al 2014). However, because data from multiple detection methods may not be independent if detectors are co-located (i.e., >1 detection method at similar spatial locations), dependence between detection methods, if present, must be accounted for to mitigate parameter estimate bias (Clare et al 2017). However, because data from multiple detection methods may not be independent if detectors are co-located (i.e., >1 detection method at similar spatial locations), dependence between detection methods, if present, must be accounted for to mitigate parameter estimate bias (Clare et al 2017).…”