PARENT SESSION
Monday, August 7, 8:00-11:30 am
COS 11 - Remote sensing
L-3, Lobby Level, Cook Convention Center
Presiders: G Carter and D Rocchini

Mapping and modeling pitcher-plant habitat using remote sensing, GIS, and expert classification.

Blossom, Gabriel^*,1, Davis, Micheal¹, Carter, Gregory², Griffith, Jerry ³, Diehl, Robert¹, ¹ The University of Southern Mississippi, Hattiesburg, MS² The University of Southern Mississippi, Ocean Springs, MS³ The University of Southern Mississippi, Hattiesburg, MS

ABSTRACT- Pitcher-plant bogs and wet pine savannas (hereafter as pitcher-plant habitat or PPH) represent one of the most diverse ecosystems in the southeastern United States. Approximately 98% of all original PPH has been destroyed throughout the southeastern coastal plain and the remaining communities are threatened. Management of species richness within these ecosystems typically involves prescribed burning to keep native woody invasive species from colonizing and displacing herbaceous plant communities. Understanding the effects of management practices on PPH and monitoring habitat loss requires the systematic collection of temporally useful data over large areas. This study examines the feasibility of using fine resolution, multispectral satellite imagery (Quickbird) in conjunction with digital elevation models (DEM) and Geographic Information System (GIS) layers to produce an accurate map of PPH within the Desoto National Forest of southern Mississippi. Supervised maximum likelihood classification was used to produce a vegetation map. This map was then further refined using a decision tree classifier that incorporates 3 GIS layers and a DEM wetness layer. The decision tree modeled potential bog habitat and corrected errors produced by spectral similarities between classes, using logic operators based on spatially explicit environmental variables. The final thematic map had an overall accuracy of K(kappa)=59.8 while the original spectral map had a accuracy of K=36.5. Our results indicated that PPH can be studied and monitored with acceptable accuracy using a hybridized GIS and remote sensing approach.

Key words: remote sensing, ecosystem management, expert classification