Continuous-time spatially explicit capture–recapture models, with an application to a jaguar camera-trap survey
- Many capture–recapture surveys of wildlife populations operate in continuous time, but detections are typically aggregated into occasions for analysis, even when exact detection times are available. This discards information and introduces subjectivity, in the form of decisions about occasion definition.
- We develop a spatiotemporal Poisson process model for spatially explicit capture–recapture (SECR) surveys that operate continuously and record exact detection times. We show that, except in some special cases (including the case in which detection probability does not change within occasion), temporally aggregated data do not provide sufficient statistics for density and related parameters, and that when detection probability is constant over time, our continuous-time (CT) model is equivalent to an existing model based on detection frequencies. We use the model to estimate jaguar density from a camera-trap survey and conduct a simulation study to investigate the properties of a CT estimator and discrete-occasion estimators with various levels of temporal aggregation. This includes investigation of the effect on the estimators of spatiotemporal correlation induced by animal movement.
- The CT estimator is found to be unbiased and more precise than discrete-occasion estimators based on binary capture data (rather than detection frequencies) when there is no spatiotemporal correlation. It is also found to be only slightly biased when there is correlation induced by animal movement, and to be more robust to inadequate detector spacing, while discrete-occasion estimators with binary data can be sensitive to occasion length, particularly in the presence of inadequate detector spacing.
- Our model includes as a special case a discrete-occasion estimator based on detection frequencies, and at the same time lays a foundation for the development of more sophisticated CT models and estimators. It allows modelling within-occasion changes in detectability, readily accommodates variation in detector effort, removes subjectivity associated with user-defined occasions and fully utilizes CT data. We identify a need for developing CT methods that incorporate spatiotemporal dependence in detections and see potential for CT models being combined with telemetry-based animal movement models to provide a richer inference framework.