May 22, 2018
May 24th RSGAL Dr. Meghan Halabisky’s talk at WAURISA
Semi-automated, remote sensing based approach for updating the National Wetland Inventory in Washington State
n Washington State, the existing statewide wetland maps (National Wetland Inventory [NWI] maps) are out of date and inaccurate in many locations. These errors of omission have been recorded to be as high as 50% in some areas, and may be as high as 90% in some forested areas. Inaccuracies and errors of omission are due in part to the difficulty of photo-interpreting certain land cover types (e.g., forested wetlands, wetlands on slopes, and vernal pools), especially when using lower spectral, spatial, and temporal resolution imagery from the 1980s. Also, many wetlands on agricultural lands were not mapped. Recent development of a suite of new tools has shown an improvement in mapping efforts of wetlands, however, most studies were tested in more homogenous landscapes compared to the rugged and diverse terrain dominated by evergreen trees found in areas like Washington State. The objectives of our research was to 1.) Compare the accuracy of two high resolution remote sensing classification techniques; Random forest classification and rule based classification, integrating object based image analysis, Lidar-derived datasets, and hydrologic flow models in two areas representative of land use and ecological diversity of WA. 2.) Determine mapping and classification error among different wetland types, geographic areas, land uses, and ecological systems (incorporating ground-truthing) in the study areas.
This abstract is part of a land cover, wetland, and change mapping session which include the following 4 authors: Nate Herold, Meghan Halbisky, Ken Pierce, and Suzanne Shull.
