Date of Graduation

5-2012

Document Type

Thesis

Degree Name

Master of Arts in Geography (MA)

Degree Level

Graduate

Department

Geosciences

Advisor/Mentor

Dixon, John C.

Committee Member

Tullis, Jason A.

Second Committee Member

Huxel, Gary R.

Keywords

Earth sciences; Biological sciences; Classification; Gis; Land cover; Mapping

Abstract

Maintaining representative sampling of biologically rich and rare ecosystems has become an important means to preventing biodiversity loss. A limitation in indentifying and quantifying ecosystems is the cost of obtaining high resolution imagery necessary for a high resolution land cover assessment. This research shows how free, different resolution imagery (orthoimages and LANDSAT ETM+) could be combined to produce a hybrid dataset with enhanced spectral, spectral and temporal properties, and processed to obtain a object-based classification of land cover of bottomland and pine hardwood forest in south eastern Arkansas. Three classification techniques were evaluated: 1) a human derived, rule based method, 2) A nearest neighbor classification using only the infrared orthoimage (SRGB), and 3) A nearest neighbor classification using the infrared orthoimage and LANDSAT ETM+ derived multitemporal NDVI values (SNDVI). Overall accuracy of the rule based method and SNDVI were comparable, and significantly higher (~10-20%) than the SRGB. Further, when compared to existing land cover maps, both the rule based method and SNDVI had far greater visual appeal and accuracy.

Download

Share

COinS