Using a remote sensing/GIS model to predict southwestern Willow Flycatcher breeding habitat along the Rio Grande, New Mexico

Abstract

Introduction The Southwestern Willow Flycatcher (Empidonax traillii extimus; hereafter SWFL) is a federally endangered bird (USFWS 1995) that breeds in riparian areas in portions of New Mexico, Arizona, southwestern Colorado, extreme southern Utah and Nevada, and southern California (USFWS 2002). Across this range, it uses a variety of plant species as nesting/breeding habitat, but in all cases prefers sites with dense vegetation, high canopy, and proximity to surface water or saturated soils (Sogge and Marshall 2000). As of 2005, the known rangewide breeding population of SWFLs was roughly 1,214 territories, with approximately 393 territories distributed among 36 sites in New Mexico (Durst et al. 2006), primarily along the Rio Grande. One of the key challenges facing the management and conservation of the Southwestern Willow Flycatcher is that riparian areas are dynamic, with individual habitat patches subject to cycles of creation, growth, and loss due to drought, flooding, fire, and other disturbances. Former breeding patches can lose suitability, and new habitat can develop within a matter of only a few years, especially in reservoir drawdown zones. Therefore, measuring and predicting flycatcher habitat - either to discover areas that might support SWFLs, or to identify areas that may develop into appropriate habitat - requires knowledge of recent/current habitat conditions and an understanding of the factors that determine flycatcher use of riparian breeding sites. In the past, much of the determination of whether a riparian site is likely to support breeding flycatchers has been based on qualitative criteria (for example, 'dense vegetation' or 'large patches'). These determinations often require on-the-ground field evaluations by local or regional SWFL experts. While this has proven valuable in locating many of the currently known breeding sites, it is difficult or impossible to apply this approach effectively over large geographic areas (for example, the middle Rio Grande). The SWFL Recovery Plan (USFWS 2002) recognizes the importance of developing new approaches to habitat identification, and recommends the development of drainage-scale, quantitative habitat models. In particular, the plan suggests using models based on remote sensing and Geographic Information System (GIS) technology that can capture the relatively dynamic habitat changes that occur in southwestern riparian systems. In 1999, Arizona Game and Fish Department (AGFD) developed a GIS-based model (Hatten and Paradzick 2003) to identify SWFL breeding habitat from Landsat Thematic Mapper imagery and 30-m resolution digital elevation models (DEMs). The model was developed with presence/absence survey data acquired along the San Pedro and Gila rivers, and from the Salt River and Tonto Creek inlets to Roosevelt Lake in southern Arizona (collectively called the project area). The GIS-based model used a logistic regression equation to divide riparian vegetation into 5 probability classes based upon characteristics of riparian vegetation and floodplain size. This model was tested by predicting SWFL breeding habitat at Alamo Lake, Arizona, located 200 km from the project area (Hatten and Paradzick 2003). The GIS-based model performed as expected by identifying riparian areas with the highest SWFL nest densities, located in the higher probability classes. In 2002, AGFD applied the GIS-based model throughout Arizona, for riparian areas below 1,524 m (5,000 ft) elevation and within 1.6 km of perennial or intermittent waters (Dockens et al. 2004). Overall model accuracy (using probability classes 1-5, with class 5 having the greatest probability of nesting activity) for predicting the location of 2001 nest sites was 96.5 percent; accuracy decreased when fewer probability classes were defined as suitable. Map accuracy, determined from errors of commission, increased in higher probability classes in a fashion similar to errors of omission. Map accuracy, li