Using occupancy models of forest breeding birds to prioritize conservation planning

Abstract

As urban development continues to encroach on the natural and rural landscape, land-use planners struggle to identify high priority conservation areas for protection. Although knowing where urban-sensitive species may be occurring on the landscape would facilitate conservation planning, research efforts are often not sufficiently designed to make quality predictions at unknown locations. Recent advances in occupancy modeling allow for more precise estimates of occupancy by accounting for differences in detectability. We applied these techniques to produce robust estimates of habitat occupancy for a subset of forest breeding birds, a group that has been shown to be sensitive to urbanization, in a rapidly urbanizing yet biological diverse region of New York State. We found that detection probability ranged widely across species, from 0.05 to 0.8. Our models suggest that detection probability declined with increasing forest fragmentation. We also found that the probability of occupancy of forest breeding birds is negatively influenced by increasing perimeter–area ratio of forest fragments and urbanization in the surrounding habitat matrix. We capitalized on our random sampling design to produce spatially explicit models that predict high priority conservation areas across the entire region, where interior-species were most likely to occur. Finally, we use our predictive maps to demonstrate how a strict sampling design coupled with occupancy modeling can be a valuable tool for prioritizing biodiversity conservation in land-use planning.

Introduction

Habitat loss and fragmentation present a significant threat to the persistence of biological communities (e.g. Andrén, 1994, Murcia, 1995, Vitousek et al., 1997). Urban and exurban development have been identified as the primary drivers of habitat loss worldwide (Brown et al., 2005, Hansen et al., 2005, Williams et al., 2005, Kavzoglu, 2008). The associated characteristics of urbanization (e.g. increased road density and edge effects) can have a dramatic impact on the abundance, extinction rate and diversity of a number of vertebrate species (Boulinier et al., 2001, Devictor et al., 2007). As such, identifying and incorporating high priority conservation areas into land-use planning will be a critical step in the long-term protection of biodiversity.

In North America, loss of breeding habitat to urbanization and increasing fragmentation has been identified as a potential cause of population declines in Neotropical migrants (Robbins et al., 1989, Parody et al., 2001; Smith and Wachob, 2006). Due to widespread decline in core habitat, forest breeding bird species may be of particular conservation concern. Numerous studies have shown that urban-sensitive species such as the scarlet tanager (Piranga olivacea), cerulean warbler (Dendroica cerulea), and Louisiana waterthrush (Seiurus motacilla) decline in landscapes with lower mean forest-patch size, and higher residential development (e.g. Blair, 1996, Boulinier et al., 2001, Groom and Grubb, 2002). Though quality data is often not readily available, knowing where urban-sensitive forest breeding species are or could be occurring in an urbanizing landscape would be important for targeting conservation actions (Peterson and Dunham, 2003).

Predictive models of species richness are becoming increasingly popular for prioritizing conservation efforts (Jewell et al., 2007). Typically, these models rely on species distribution data (Hijmans and Graham, 2006), inferred patterns from remotely-sensed habitats (Scott et al., 1993), inventories (Müller et al., 2003), or occupancy (presence/absence) data (Hall and Schafale, 2009). However, these approaches are often not based on probability sampling and, as a consequence, the resulting data may be unreliable (Conroy and Noon, 1996). For forest breeding birds, predictive mapping of species richness can be hindered by a number of technical and ecological challenges. For robust predictions, sample size and design requirements can limit the practicality and value of spatial models (Chatfield, 1995). In addition, many avian studies rely heavily on surveys that occur on or near roads (e.g. Robbins et al., 1989, Somershoe, 2006), though the relative composition of habitat along roads can be significantly different than what is available across the species’ range (Bart et al., 1995). Finally, although high concentrations of important species are expected in regions where appropriate habitat is available and contiguous (e.g. Taulman and Smith, 2002), large forest tracts are becoming increasingly rare in more urbanized landscapes. As such, predicting regions of significant bird diversity based on habitat alone may be challenging. These factors make predicting species diversity in more urbanized landscapes a complex, unreliable, or even impossible tool for conservation.

Occupancy is often used as an inexpensive surrogate, or state variable, to predict habitat-use for individual and groups of species at unknown locations (e.g. Hall and Schafale, 2009, Kavanagh and Stanton, 2005). However, research has shown that estimates of occupancy that do not account for the inability to detect all species during sampling are negatively biased (Nichols et al., 1998, MacKenzie et al., 2006, MacKenzie, 2006). An individual may go undetected in a unit that is actually being used if it fails to call or leave signs that are visible or audible to the observer (MacKenzie, 2006). Habitat-use models can also be misinterpreted when detection is a function of the landscape covariates that are expected to affect occupancy (MacKenzie, 2006). Naïve estimates of presence/absence tend to overestimate the strength of the relationship between habitat covariates that are associated with detection, and occupancy (e.g. Bailey et al., 2004, MacKenzie, 2006, Hossack and Corn, 2007). In order for habitat-use maps to be a more valuable and reliable tool for conservation and management, the impact of detection on occupancy must be explicitly incorporated.

Recent advances in occupancy modeling provide a framework and software for dealing with imperfect detection (MacKenzie et al., 2006, Bailey et al., 2007). The data required for these models are similar to typical presence/absence designs. However, multiple surveys need to be conducted at each sample unit within a specified time period (detection history) to account for individuals who are there, but not detected (MacKenzie, 2006). If applied to habitat-use models, managers would be more confident in predicting where species could be occurring on the landscape, and which high-use areas should be targeted for conservation.

The purpose of our study was to identify high priority conservation areas based on the probability of occupancy for nine forest breeding bird species in a rapidly urbanizing region of New York State, USA. We used a strict random sampling design, coupled with advances in modeling techniques, to create spatially explicit predictive maps of habitat occupancy. A greater probability of occupancy indicated a habitat that was more likely to be used by these species. We also examined the relative influence of habitat availability and fragmentation on both occupancy and detection probability. Finally, we applied these models to a real-world conservation problem by demonstrating how occupancy models could be used to target targeting outreach and education for land-use planners.