Multi-temporal Sub-pixel Landsat ETM plus Classification of Isolated Wetlands in Cuyahoga County, Ohio, USA

The goal of this research was to determine the utility of subpixel processing of multi-temporal Landsat Enhanced Thematic Mapper Plus (ETM+) data for the identification and mapping of isolated wetlands a parts per thousand yen 0.20 ha (0.50 acres) in Cuyahoga County, Ohio. Segmentation and object-oriented analysis of Landsat ETM+ was used to map forested and emergent marsh isolated wetlands in Alachua County, Florida, previously; however, the isolated wetlands in our study area lacked the well-defined, high-contrast boundaries between wetland and surrounding upland needed to make this method successful. We developed a new methodology that incorporated Landsat ETM+; a Normalized Difference Vegetation Index mask; and subpixel matched filtering-which determines the apparent abundance of wetlands at subpixel levels in the presence of spectrally-mixed, unknown background through a partial unmixing algorithm-to map > 43 km(2) (16 mi(2)) of isolated wetlands in our 1,189 km(2) (459 mi(2)) study area. The final overall accuracy of the classification was 92.8%, with a Kappa coefficient of 0.86; producer accuracy for isolated wetlands was 87.9% (omission error 12.1%) and user accuracy was 97.4% (commission error 2.6%). The subpixel matched filtering method used in this research appears to provide an effective means for mapping isolated wetlands a parts per thousand yen 0.20 ha, especially those with boundaries that are not easily identified.