Detecting population decline of birds using long-term monitoring data

Abrupt population change in birds may be caused by various factors. When such events occur, it is important to understand the population-level impact on the species. We applied a change point analysis with Markov chain Monte Carlo using long-term population count data to address this question. We first investigated the method with a simple Poisson model using synthetic data sets for different population decline scenarios and number of observations. Estimated change points were particularly accurate when a large decline in counts occurred. Accuracy and precision of posterior change magnitude tended to increase when actual change magnitude became larger. We applied the method to two cases using data from the North American Breeding Bird Survey: epidemic mortality of Florida scrub-jays (Aphelocoma coerulescens) in central Florida and population decline of American crows (Corvus brachyrhynchos) in Maryland and Virginia after West Nile virus emergence. The Florida scrub-jay case study indicated that the estimated change point was accurate compared with that reported by local monitoring. A Poisson-log model that included observer and year variability resulted in better fit to the data than a simple Poisson model. The American crow case study showed that the method detected change points towards the end of observational data, but not all change point parameters converged, which may suggest that a population decline did not occur or was small for some survey routes we analyzed. Our study demonstrates the utility of change point analysis to examine abrupt population change. Data from systematic long-term monitoring can be a basis of such an analysis.