A generalized observation confirmation model to account for false positive error in species detection-nondetection data

Spatially-indexed repeated detection-nondetection data is widely collected by ecologists interested in estimating parameters associated with species distribution, relative abundance, phenology, and more while accounting for imperfect detection. Recent model development has focused on accounting for false positive error as well, given growing recognition that misclassification is common across many sampling protocols. To date, however, the development of model-based solutions to false positive error has been largely restricted to occupancy models. We describe a general form of the observation confirmation protocol originally described for occupancy estimation that permits investigators to flexibly and intuitively extend several models for detection-nondetection data to account for false positive error. Simulation results demonstrate that estimators for relative abundance and arrival time exhibit relative bias greater than 20% under realistic levels of false positive prevalence (e.g., 5% of detections are false positive). Bias increases as true and false positives occur in more distinct places or times, but can also be sensitive to the values of the state variables of interest, sampling design, and sampling efficiency. Results from an empirical study focusing on patterns of gray fox relative abundance across Wisconsin, USA suggest that false positive error can also distort estimated spatial patterns often used to guide decision-making. The extended estimators described within typically improve performance at any level of confirmation, and when false positive error occurs at random and constitutes less than 10% of all detections, the estimators are essentially unbiased when more than 50 observations can be confirmed as true or false positives. The generalized form of the observation-confirmation protocol is a flexible model-based solution to false positive error useful for researchers collecting data with sampling devices like trail or smartphone cameras, acoustic recorders, or other techniques where classifications can be reviewed post-hoc.