| Abstract Detail
Ecological Section Bourg, Norman A. [1], McShea, William J. [2], Gill, Douglas E. [1]. Classification Tree and GIS-based Predictive Habitat Modeling for Xerophyllum asphodeloides (L.) Nutt. (Melanthiaceae), a Rare Fire-adapted Appalachian Forest Herb. We used a classification tree (CART) model in combination with GIS to predict suitable habitat and potential new occurrences for Turkeybeard (Xerophyllum asphodeloides), a rare lily associated with southern Appalachian pine-oak forests. Turkeybeard is included in the US Center for Plant Conservation's National Collection of Endangered Plants. We gathered evidence from field experiments and long-term population monitoring supporting the hypothesis that turkeybeard is a fire-adapted species, relying on this disturbance for long-term population maintenance. Based upon these findings, we compiled fire occurrence records for three National Forest districts and then performed kriging interpolation on a subset of these data to construct a fire likelihood GIS layer. GIS data layers describing forest and soil types, elevation, slope, aspect, and planar and profile landform indices were also compiled for the study area. We used population survey and sample data from each layer in our model to produce a cross-validated classification tree that predicted suitable habitat in the study area. Fire likelihood was one of four main explanatory variables in the model. Approximately 4% of the study area was classified into five suitable habitat classes, with a misclassification error rate of 4.74%. The model correctly classified 74% and 90% of the known presence and absence areas respectively, and ground-truthing surveys discovered eight new occupied habitat patches. Six false negatives were found, but these all occurred near predicted suitable habitat harboring known or new populations. This study’s implications are important not only for conservation and management of X. asphodeloides, but also as confirmation of the potential and value of CART/GIS-based modeling approaches to species distribution problems in ecology. Future work includes combining the model with data from a hyperspectral remote sensor to better define occupied habitat, and comparing population genetic and spatial structure in this species at the metapopulation level.
1 - University of Maryland, Dept. of Biology, 1204 Biology-Psychology Building, College Park, Maryland, 20742, USA 2 - Smithsonian Institution, National Zoological Park, Conservation and Research Center, 1500 Remount Road, Front Royal, Virginia, 22630, U.S.A.
Keywords: classification tree Ecology Appalachian Mountains geographical distribution Melanthiaceae predictive model rare taxa.
Presentation Type: Poster Session: 32-39 Location: Special Event Center (Cliff Lodge) Date: Tuesday, August 3rd, 2004 Time: 12:30 PM Abstract ID:151 |