A satellite model of Southwestern Willow Flycatcher ( Empidonax traillii extimus ) breeding habitat and a simulation of potential effects of tamarisk leaf beetles ( Diorhabda spp.), southwestern United States

Abstract

Executive Summary The study described in this report represents the first time that a satellite model has been used to identify potential Southwestern Willow Flycatcher ( Empidonax traillii extimus ) (hereinafter referred to as “flycatcher”) breeding habitat rangewide for 2013–15. Fifty-seven Landsat scenes were required to map the entire range of the flycatcher, encompassing parts of six States and more than 1 billion 30-meter pixels. Predicted flycatcher habitat was summarized in a hierarchical fashion from largest to smallest: regionwide, State, U.S. Fish and Wildlife Service (FWS) management unit, 7.5-minute quadrangle, and critical-habitat reach. The term “predicted habitat” is used throughout this report to distinguish areas the satellite model predicts as suitable flycatcher habitat from what may actually exist on the ground. A rangewide accuracy assessment was done with 758 territories collected in 2014, and change detection was done with yearly habitat maps to identify how and where habitat changed over time. Additionally, effects of tamarisk leaf beetles ( Diorhabda spp.) on flycatcher habitat were summarized for the lower Virgin River from 2010 to 2015, and simulations of how tamarisk leaf beetles may affect flycatcher habitat in the lower Colorado and upper Gila Rivers were done for 2015. Model results indicated that the largest areas of predicted flycatcher habitat at elevations below 1,524 meters were in New Mexico and Arizona, areas followed in descending order by California, Texas, Nevada, Utah, and Colorado. By FWS management unit, the largest area of flycatcher habitat during all 3 years were the Middle Rio Grande (New Mexico), followed by the Upper Gila (Arizona and New Mexico) and Middle Gila/San Pedro (Arizona) management units. The area of predicted flycatcher habitat varied considerably in 7.5-minute quadrangles, ranging from 0 to1,398 hectares (ha). Averaged across 3 years, the top three producing quadrangles were Paraje Well (New Mexico), San Marcial (New Mexico), and San Carlos Reservoir (Arizona). The top three FWS critical-habitat reaches in 2015 were Rio Grande-middle (9,544 ha), San Pedro River (1,779 ha), and Gila River-mid San Carlos (1,356 ha); this ranking did not change in 2013 or 2014. Change detection among years showed a large shift in predicted flycatcher habitat influenced by drought patterns, with California habitat decreasing and New Mexico habitat increasing. An accuracy assessment indicated that 88 percent of territories were correctly classified at a 40 percent probability threshold, with an exponential relationship between territory densities and five probability classes. A spatially explicit analysis indicated that beetles decreased predicted flycatcher habitat 94.2 percent from 2010 to 2015 along the lower Virgin River, with only 5.8 percent persisting. In contrast, beetle simulations indicated that 64.1 percent of habitat will persist along the lower Colorado River and 45 percent will persist along the upper Gila River. This project shows that the satellite model adequately predicts flycatcher habitat rangewide, but it lacks the ability to predict which patches will be occupied in a given year. The next logical step is the development of an occupancy model that ties the habitat predictions of the satellite model to patch occupancy so managers can better allocate their resources for survey and restoration activities. Finally, the methods presented in this report seem well suited for automated mapping applications and cloud-based resources.