Dealing with incomplete and variable detectability in multi-year, multi-site monitoring of ecological populations

An ecological monitoring program should be viewed as a component of a larger framework designed to advance science and/or management, rather than as a stand-alone activity. Monitoring targets (the ecological variables of interest; e.g. abundance or occurrence of a species) should be set based on the needs of that framework (Nichols and Williams 2006; e.g. Chapters 2–4). Once such monitoring targets are set, the subsequent step in monitoring design involves consideration of the field and analytical methods that will be used to measure monitoring targets with adequate accuracy and precision. Long-term monitoring programs will involve replication of measurements over time, and possibly over space; that is, one location or each of multiple locations will be monitored multiple times, producing a collection of site visits (replicates). Clearly this replication is important for addressing spatial and temporal variability in the ecological resources of interest (Chapters 7–10), but it is worth considering how this replication can further be exploited to increase the effectiveness of monitoring. In particular, defensible monitoring of the majority of animal, and to a lesser degree plant, populations and communities will generally require investigators to account for imperfect detection (Chapters 4, 18). Raw indices of population state variables, such as abundance or occupancy (sensu MacKenzie et al. 2002), are rarely defensible when detection probabilities are < 1, because in those cases detection may vary over time and space in unpredictable ways. Myriad authors have discussed the risks inherent in making inference from monitoring data while failing to correct for differences in detection, resulting in indices that have an unknown relationship to the parameters of interest (e.g. Nichols 1992, Anderson 2001, MacKenzie et al. 2002, Williams et al. 2002, Anderson 2003, White 2005, Kéry and Schmidt 2008). While others have argued that indices may be preferable in some cases due to the challenges associated with estimating detection probabilities (e.g. McKelvey and Pearson 2001, Johnson 2008), we do not attempt to resolve this debate here. Rather, we are more apt to agree with MacKenzie and Kendall (2002) that the burden of proof ought to be on the assertion that detection probabilities are constant. Furthermore, given the wide variety of field methods available for estimating detection probabilities and the inability for an investigator to know, a priori, if detection probabilities will be constant over time and space, we believe that development of monitoring programs ought to include field and analytical methods to account for the imperfect detection of organisms.