A Geographic Ontology and GIS Model for Carolina Bays

ASU Author/Contributor (non-ASU co-authors, if there are any, appear on document)
Jacob Richmond Turner (Creator)
Institution
Appalachian State University (ASU )
Web Site: https://library.appstate.edu/
Advisor
Christopher Badurek

Abstract: Carolina bays are a unique geomorphologic entity located along the Atlantic coastal plain. Even without the benefit of an overhead view, they have been noted as a distinct feature of the coastal plain as first described by the South Carolina Geological Survey of South Carolina in 1848. The first aerial photographs in the 1930 coastal South Carolina region revealed that the unique depression wetlands were more than just a strange local phenomenon. Aerial photos enabled observers to see qualities in addition to their relative distribution that make them unique: their oval shape, northwest to southeast orientation and the presence of raised sand rims along their eastern and southeastern edges in many instances. Being such a distinctive surface feature and recognized for their ecological value, it would seem that Carolina bays would have been defined within their own map coverage across the Atlantic Coastal plain. However, just two statewide inventories have been completed for South Carolina and Georgia, and one for North Carolina has never been conducted. While previous inventories have employed onscreen digitization with Geographic Information Systems (GIS) in order to inventory bays, researchers have raised concerns over how individuals define Carolina bay as a geographic entity. The differences in human perception make the classification of geographic entities that exist on a continuum such as Carolina bays a challenge and may have contributed to widely varying estimates of their numbers. In order to explore the classification issues related to Carolina bays, and the usefulness of geographic ontology and cartographic modeling for inventory, a cartographic model was constructed for use within the Ocean Bay quad in Francis Marion National Forest in Berkeley and Charleston Counties, South Carolina. To test the model’s selective ability, a comparison was made between Carolina bay features that a researcher selected and bays identified by a cartographic model. The model positively identified 76 percent of Carolina bays that a researcher identified in an image within a single quadrangle. The approach used in this model showed that the initial identification rule of any pixel within a bay’s border counted as a positive identification was inadequate. Other aspects not accounted for, including false positive identification, neither researcher nor model being able to identify a bay, or bays that the model was able to select that the researcher was not were added into a subsequent model. Results from the amended model show fewer researcher identified instances of Carolina bays, but a slightly higher rate of mutual identification by the model and the researcher. With these complications in mind, a similar approach was taken with Bladen County North Carolina, but with significant revisions. A cartographic model was created for Bladen County North Carolina in which bay characteristics were selected from the North Carolina Gap Analysis Program (GAP) land use/landcover dataset, the Soil Survey Geographic Database (SSURGO) and National Wetlands Inventory (NWI). The predictive ability of the model was assessed by manually selecting Carolina bays from a high resolution image and comparing the manually selected bays with the model identifications. In order to remedy the issue of forcing all instances of bays into one of two categories (either an object is a bay or it is not), a ranking system was developed that was based upon a core/radial cognitive model, and the approach taken with the Savannah River Ecology Lab (SREL) inventory. The rule for positive identification was changed from a single pixel to a visual estimation of 50 percent coverage of a Carolina bay. As a whole, the predictive model identified 57 percent of the features also identified manually by the researcher, but the bay ranking system gives a different breakdown of how well the model worked within each category: exemplar (86 percent), less distinct (79 percent), bay-like (53 percent), and destroyed (19 percent) show significant differences. In addition to the ranking system, other attributes were assessed, such as the presence or absence of a sand rim, water visibility, overlap, diverging, long axis length and orientation. The analysis shows that the model has the potential to identify well defined bays with at least 50 percent areal coverage, and as such offers the first iteration of a computational ontology for the Carolina bays of Bladen County, North Carolina. Results from this research may provide a basis for modeling the entire range of Carolina bays, defining one of the most curious features of the Atlantic Coastal Plain and uniting differing definitions under one digital concept.

Additional Information

Publication
Thesis
Tuner, J.R. (2010). A Geographic Ontology and GIS Model for Carolina Bays. Unpublished master's thesis. Appalachian State University, Boone, NC
Language: English
Date: 2010

Email this document to