Show simple item record

dc.contributor.authorHladik, Christine Michelle
dc.date.accessioned2014-03-04T20:36:00Z
dc.date.available2014-03-04T20:36:00Z
dc.date.issued2012-08
dc.identifier.otherhladik_christine_m_201208_phd
dc.identifier.urihttp://purl.galileo.usg.edu/uga_etd/hladik_christine_m_201208_phd
dc.identifier.urihttp://hdl.handle.net/10724/28291
dc.description.abstractSalt marshes are valuable ecosystems that are susceptible to habitat loss due to changes in sea level and coastal flooding, and there is growing interest in obtaining accurate habitat and elevation maps for these areas. Remote sensing techniques such as Light Detection and Ranging (LIDAR) can produce digital elevation models (DEMs), but the accuracy of LIDAR in salt marshes is limited by a combination of sensor resolution, instrument errors, and poor laser penetration in dense vegetation. I assessed the accuracy of a LIDAR-derived DEM for the salt marshes surrounding Sapelo Island, GA using real time kinematic (RTK) GPS. These observations were used to develop and validate species-specific correction factors for ten marsh cover classes, which ranged from 0.03 to 0.25 m. In order to apply these corrections to the 13 km2 study site, I classified hyperspectral imagery by cover class and combined this information with elevation in a decision tree. This produced both an accurate habitat classification (nine salt marsh habitat classes were mapped with a 90% overall accuracy) and a corrected DEM (overall mean error was reduced from 0.10 ± 0.12 (SD) to -0.003 ± 0.10 m (SD) and root mean squared error at the 68 % confidence level decreased from 0.15 to 0.10 m) when validated with ground truth data. Finally, I evaluated the use of remote sensing-derived variables (DEM elevation, slope, distance metrics) versus field collected edaphic variables (soil organic matter, water content, salinity, redox) to develop predictive models of plant distributions with both linear discriminant analysis (LDA) and classification and regression trees (CART). Models that used remote sensing variables had accuracies of 0.78 and 0.79, whereas those for edaphic models were 0.63 and 0.72 for LDA and CART, respectively. Accuracies improved only slightly in the best models which combined remote sensing variables and soil organic matter (to 0.82 and 0.83 for LDA and CART, respectively), suggesting that remote sensing-derived variables alone can be effective predictors of marsh vegetation. Taken together, these findings show the potential for appropriately analyzed remote sensing data for evaluating elevation and habitat in marshes.
dc.languageeng
dc.publisheruga
dc.rightspublic
dc.subjectRemote sensing
dc.subjectLIDAR
dc.subjectdigital elevation model (DEM)
dc.subjecthyperspectral imagery
dc.subjectsalt marsh
dc.subjecthabitat mapping
dc.subjectlinear discriminant analysis (LDA)
dc.subjectclassification and regression trees (CART)
dc.subjectSapelo Island
dc.subjectLTER
dc.titleUse of remote sensing data for evaluating elevation and plant distribution in a southeastern salt marsh
dc.typeDissertation
dc.description.degreePhD
dc.description.departmentMarine Sciences
dc.description.majorMarine Sciences
dc.description.advisorMerryl Alber
dc.description.committeeMerryl Alber
dc.description.committeeSteven C. Pennings
dc.description.committeeJames Morris
dc.description.committeeMarguerite Madden
dc.description.committeeClark Alexander


Files in this item

FilesSizeFormatView

There are no files associated with this item.

This item appears in the following Collection(s)

Show simple item record