Quantitative analysis of woodpecker habitat using high-resolution airborne LiDAR estimates of forest structure and composition

Authors

James E. Garabedian, North Carolina State UniversityFollow
Robert McGaughey, Pacific Northwest Research StationFollow
Stephen E. Reutebuch, Pacific Northwest Research Station, USDA Forest ServiceFollow
Bernard R. Parresol, Southern Research Station, USDA Forest ServiceFollow
John C. Kilgo, Southern Research Station, USDA Forest ServiceFollow
Christopher E. Moorman, North Carolina State UniversityFollow
M. Nils Peterson, North Carolina State UniversityFollow

Document Type

Article

Date of this Version

2014

Citation

Remote Sensing of Environment 145 (2014), pp. 68–80; doi: 10.1016/j.rse.2014.01.022

Comments

This article is a U.S. government work, and is not subject to copyright in the United States.

Abstract

Light detection and ranging (LiDAR) technology has the potential to radically alter theway researchers and managers collect data onwildlife–habitat relationships. To date, the technology has fostered several novel approaches to characterizing avian habitat, but has been limited by the lack of detailed LiDAR-habitat attributes relevant to species across a continuum of spatial grain sizes and habitat requirements. We demonstrate a novel three-step approach for using LiDAR data to evaluate habitat based on multiple habitat attributes and accounting for their influence at multiple grain sizes using federally endangered red-cockaded woodpecker (RCW; Picoides borealis) foraging habitat data fromthe Savannah River Site (SRS) in South Carolina, USA. First,we used high density LiDAR data (10 returns/m2) to predict detailed forest attributes at 20-mresolution across the entire SRS using a complementary application of nonlinear seemingly unrelated regression andmultiple linear regressionmodels. Next,we expanded on previous applications of LiDAR by constructing 95% joint prediction confidence intervals to quantify prediction error at various spatial aggregations and habitat thresholds to determine a biologically and statistically meaningful grain size. Finally,we used aggregations of 20-m cells and associated confidence interval boundaries to demonstrate a newapproach to produce maps of RCWforaging habitat conditions based on the guidelines described in the species' recovery plan. Predictive power (R2) of regression models developed to populate raster layers ranged from 0.34 to 0.81, and prediction error decreased as aggregate size increased, but minimal reductions in prediction error were observed beyond 0.64-ha (4 × 4 20-m cells) aggregates. Mapping habitat quality while accounting for prediction error provided a robust method to determine the potential range of habitat conditions and specific attributes that were limiting in terms of the amount of suitable habitat. The sequential steps of our analytical approach provide a useful framework to extract detailed and reliable habitat attributes for a forest-dwelling habitat specialist, broadening the potential to apply LiDAR in conservation and management of wildlife populations.

A zipped folder of Google maps is attached below as a related file.